# Lily AI — full content > Lily AI is the product intelligence company behind Lily Max, an agentic engine that enriches retail product catalogs and proves lift with controlled tests across Google Ads, Meta Ads, AI shopping assistants and onsite search. This is the expanded companion to /llms.txt: the full text of the company page, the product pages, the topic hubs, every long-form guide, every customer story, and the FAQ hub on https://www.lily.ai. --- # About Lily AI https://www.lily.ai/about Lily AI is the product intelligence company behind Lily Max, an agentic engine that enriches retail product catalogs and proves lift with controlled tests across Google Ads, Meta Ads, AI shopping assistants and onsite search. Founded 2015. Legal name: Lily AI, Inc.. Purva Gupta is the co-founder and ceo. Purva Gupta is the co-founder and CEO of Lily AI, the product intelligence company behind Lily Max. She writes about product data, controlled testing, and how AI surfaces decide which products shoppers see. --- # How Lily works https://www.lily.ai/how-it-works Lily ingests your catalog and performance data, builds an AI model of your products and shoppers, and runs optimized campaigns across channels. ## 1. Connect your data Link your product catalog, ad accounts, and storefront. ## 2. Lily builds understanding Lily models product attributes, demand, and shopper intent. ## 3. Launch and optimize Campaigns run and optimize continuously toward your targets. ## FAQ **How does Lily Max work?** Lily Max runs a continuous optimization loop. Your team sets the goal, then agents discover product-data gaps, generate enrichments, launch controlled tests, measure the lift, and feed the winners back into the next round, with your approval before anything ships. **Who sets the goals, the agents or my team?** Your team does. Lily Max is a goal-based system, not a black box: you define the surface, the product set, the success metric, the test cadence, and the approval rules, and the agents do the operational work toward that goal. **Do I stay in control of what gets published?** Yes. Nothing goes live without your sign-off. You review the proposed changes and the projected impact, then approve what reaches the feed, so agents handle the volume while humans own the decisions. **What are the steps in the optimization loop?** There are five: detect gaps, generate enrichments, launch a controlled test, measure the lift, and recommend what to scale. Each test makes the next one smarter, because results feed back into the system. **How does Lily Max detect product-data gaps?** Lily Max scores your catalog against what each surface rewards, then ranks the highest-impact gaps. It scans feeds, product pages, schema, and attributes for missing fields, weak titles, and thin descriptions before any enrichment work begins. **How does Lily Max prove the lift is real?** Every result is measured against a control. Lily Max uses matched-spend A/B tests, holdout groups, and difference-in-differences analysis, then reports lift with confidence intervals, so results separate validated impact from directional movement and hold up to a finance review. **What is a matched-spend A/B test?** A matched-spend A/B test runs the enriched group and the control group at the same spend level, so any difference in performance reflects the product-data change rather than a bigger budget. It is how Lily Max isolates the effect of enrichment. **How fast can Lily Max support a new AI surface?** Lily Max is built to ship support for new surface capabilities in days, not quarters. When a platform introduces a new product-data type, a catalog that is always scored and always testable can adopt it quickly, rather than waiting a release cycle. **What is the quality score, and is the score the outcome?** The score is not the outcome. Lily Max scores product information on completeness, correctness, compliance, relevance, and differentiation, and uses that score to find the signals most likely to move visibility, conversion, and ROAS, which it then tests. **How long until I see results?** Most teams connect their data and launch within days, then see results as the first controlled tests resolve. Timelines vary by catalog size and starting feed quality, and Lily Max shows where to expect impact first before any credits are spent. --- # Pricing that scales with your growth https://www.lily.ai/pricing Lily is priced to match how you grow. Because the right plan depends on your catalog size, the channels you run, and your monthly ad spend, we tailor a quote rather than list one-size-fits-all tiers. Every engagement includes the same core: agentic product-data enrichment, controlled matched-spend testing, and reporting that proves lift against a control before you scale. Below is how the plans differ, what's always included, and answers to the questions teams ask most before they book a demo. ## Starter For growing brands getting started with AI retail media on a single primary channel. Includes catalog ingestion, product-data scoring, enrichment on your highest-impact gaps, and matched-spend testing so early wins are measured, not assumed. Best when you want proof of lift before expanding. Contact us for a tailored quote. ## Growth For multi-channel brands scaling spend across Google Ads, Meta Ads, and onsite. Adds continuous optimization loops, broader catalog coverage, and cross-channel testing cadence, with human approval before anything ships to your feeds. Pricing scales with catalog size and managed spend. Contact us for a tailored quote. ## Enterprise For large retailers and marketplaces with complex catalogs and governance needs. Adds dedicated support, custom SLAs, security and compliance review, and readiness for emerging AI-discovery and agentic-commerce surfaces. Custom pricing built around your channels, catalog scale, and integration requirements. ## FAQ **How is Lily priced?** Pricing is tailored to your catalog size, the channels you run, and your monthly ad spend, rather than a fixed public rate card. This keeps the plan matched to the value Lily can move for your business. Book a demo and we'll scope a quote for your setup. **Why don't you publish fixed prices?** Results and effort scale with catalog complexity and channel mix, so a flat price would either overcharge smaller brands or under-serve large retailers. A tailored quote means you pay in line with the scope and the lift Lily is set up to deliver. **Is there a free trial or pilot?** Most teams start with a scoped pilot on their highest-impact channel. Because every plan includes matched-spend testing, the pilot itself produces measured lift against a control, so you can evaluate real impact before committing to a broader rollout. **What's included in every plan?** All plans include catalog ingestion, product-data scoring, agentic enrichment of your highest-impact gaps, matched-spend A/B testing, and reporting that isolates lift against a control. Higher tiers add channels, optimization cadence, support, and governance. **Does pricing change as I add channels?** Yes. Pricing scales with the channels you run and the spend Lily manages, so you can start on one surface and expand to Google Ads, Meta Ads, onsite, and AI-discovery surfaces as the tested lift justifies it. **Do I need to replatform or replace my current tools?** No. Lily improves the quality of the product data your existing feed managers and ad tools already send to each platform. There is no replatforming, and your team approves changes before they reach any live feed. --- # Product feed optimization https://www.lily.ai/topics/product-feed-optimization Product feed optimization is the work of improving the product-language record (titles, attributes, descriptions and taxonomy) that Google Shopping, Meta Advantage+, onsite search and AI assistants all read before deciding whether to show your product. ## Posts - What actually goes in Google Merchant Center conversational attributes (https://www.lily.ai/blog/merchant-center-conversational-attributes) - Merchant Center suspended for misrepresentation: the checklist, in the order we find them (https://www.lily.ai/blog/merchant-center-misrepresentation-suspension) - Product feed optimization: what it is, what actually moves revenue, and how to prove it (https://www.lily.ai/blog/product-feed-optimization) - Google says conversational attributes are optional. History says otherwise. (https://www.lily.ai/blog/conversational-attributes-optional-history-says-otherwise) - It always turns out to be a feed (https://www.lily.ai/blog/it-always-turns-out-to-be-a-feed) - Automation commoditized everything except your feed (https://www.lily.ai/blog/automation-commoditized-everything-except-your-feed) - $80M in incremental revenue last month, one optimization: the feed (https://www.lily.ai/blog/80m-incremental-revenue-one-optimization-the-feed) ## Case studies - How a fast-growing performance apparel brand turned product data into a 7.3% revenue lift on Google Shopping (https://www.lily.ai/case-studies/performance-apparel-google-shopping-revenue-lift) - How a global home furnishings retailer grew Google Shopping revenue 5.7%, while clicks went down (https://www.lily.ai/case-studies/home-furnishings-google-shopping-revenue-lift) - How A.L.C. boosted full-catalog product discovery on Google (https://www.lily.ai/case-studies/alc-full-catalog-discovery-on-google) - How a luxury home brand boosted sales from Shopify site search and Google search (https://www.lily.ai/case-studies/luxury-home-brand-shopify-and-google) - Tapestry boosts traffic and sales from Google Search with Lily AI (https://www.lily.ai/case-studies/tapestry-google-search-discovery) --- # Product data: attributes, taxonomy and enrichment https://www.lily.ai/topics/product-data Product data is the structured record of attributes, taxonomy and enriched descriptions that makes a catalog legible to the search engines, ad platforms and AI agents that decide what shoppers see. ## Posts - Product attributes: which ones each discovery surface reads, and what a good one looks like (https://www.lily.ai/blog/product-attributes-101) - Product Data Enrichment: What It Means When the Machines Are the Reader (https://www.lily.ai/blog/product-data-enrichment) - Product taxonomy: where a product sits, and which queries it is eligible for (https://www.lily.ai/blog/product-taxonomy) - Google just told the market AI content isn't the moat. Structured product data is. (https://www.lily.ai/blog/google-ai-content-isnt-the-moat-structured-data-is) - Brands are paying for bad product data twice now (https://www.lily.ai/blog/brands-pay-for-bad-product-data-twice) - What is Product Content Optimization? 3 Things to Know (https://www.lily.ai/blog/what-is-product-content-optimization-3-things-to-know) - From Y3K to Indie Sleaze: Brat Summer’s 4 Key Fashion Trends (https://www.lily.ai/blog/brat-summer-fashion-trends-indie-sleaze-y3k) - 2024 Trends: Macro vs. Micro (https://www.lily.ai/blog/2024-trends-macro-vs-micro) - Quiet Luxury Part Deux (https://www.lily.ai/blog/quiet-luxury-part-deux) - The Rise of Quiet Luxury (https://www.lily.ai/blog/quiet-luxury) - Lily AI’s Fashion Product Taxonomy: What is Dark Academia? (https://www.lily.ai/blog/lily-ais-fashion-product-taxonomy-what-is-dark-academia) - Lily AI’s Fashion Product Taxonomy: What is Barbiecore? (https://www.lily.ai/blog/lily-ais-fashion-product-taxonomy-what-is-barbiecore) - The Top Home Decor Styles (https://www.lily.ai/blog/the-top-home-decor-styles) ## Case studies - thredUP increases sell-through by 15% with enriched product data (https://www.lily.ai/case-studies/thredup-sell-through-15-percent) - A global luxury retailer increases relevant results for descriptive searches by up to 30x (https://www.lily.ai/case-studies/luxury-retailer-descriptive-search-30x) - A large multi-brand retailer boosts product discovery, projecting a $22M revenue increase (https://www.lily.ai/case-studies/multi-brand-retailer-22m-product-discovery) - Lily AI recommendations increase RPV by 0.65%, driving $20-30M in incremental demand (https://www.lily.ai/case-studies/multi-brand-retailer-rpv-recommendations) - How Bloomingdale's makes site search convert with enriched product data (https://www.lily.ai/case-studies/bloomingdales-site-search-convert) --- # AI discovery and agentic commerce https://www.lily.ai/topics/ai-discovery AI discovery and agentic commerce cover the surfaces where an AI assistant or agent, rather than a person, reads product data to decide which products to recommend, compare or buy. ## Posts - Agentic commerce: what it is, what changed in 2026, and what a brand actually has to do (https://www.lily.ai/blog/agentic-commerce) - How AI Shopping Assistants Actually Pick Which Products to Recommend (https://www.lily.ai/blog/how-ai-shopping-assistants-pick-products) - Stripe is building the rails. Your product data decides who wins on them. (https://www.lily.ai/blog/stripe-rails-product-data-decides-who-wins) - The Role of Product Content in AI Search and Discovery (https://www.lily.ai/blog/the-role-of-product-content-in-ai-search-and-discovery) - The Readiness Gap: Why 70% of Shoppers Are Ahead of Retailers in AI Commerce (https://www.lily.ai/blog/the-readiness-gap-why-70-of-shoppers-are-ahead-of-retailers-in-ai-commerce) - Product Discovery Disrupted: From SEO to GEO and Beyond (https://www.lily.ai/blog/product-discovery-disrupted-from-seo-to-geo-and-beyond) - The Search for Smarter Shopping: The Power of Product Content (https://www.lily.ai/blog/the-search-for-smarter-shopping-the-power-of-product-content) ## Case studies - How a global prestige beauty company turned product content into cross-channel performance and AI-recommendation visibility (https://www.lily.ai/case-studies/prestige-beauty-cross-channel-ai-visibility) - From SEM to GEO: how Merrell connects with Gen Z and millennials (https://www.lily.ai/case-studies/merrell-sem-to-geo) --- # Proof: controlled tests and case studies https://www.lily.ai/topics/proof Proof means measured lift against a control: matched-spend A/B tests and holdouts that separate what a change to product data caused from what would have happened anyway. ## Posts - Does feed optimization increase sales? The anatomy of a controlled test (https://www.lily.ai/blog/does-feed-optimization-increase-sales) - A buying standard for AI commerce (https://www.lily.ai/blog/a-buying-standard-for-ai-commerce) - The most expensive thing in retail right now (https://www.lily.ai/blog/the-most-expensive-thing-in-retail-right-now) ## Case studies - How a fast-growing performance apparel brand turned product data into a 7.3% revenue lift on Google Shopping (https://www.lily.ai/case-studies/performance-apparel-google-shopping-revenue-lift) - How a global prestige beauty company turned product content into cross-channel performance and AI-recommendation visibility (https://www.lily.ai/case-studies/prestige-beauty-cross-channel-ai-visibility) - How a global home furnishings retailer grew Google Shopping revenue 5.7%, while clicks went down (https://www.lily.ai/case-studies/home-furnishings-google-shopping-revenue-lift) - How A.L.C. boosted full-catalog product discovery on Google (https://www.lily.ai/case-studies/alc-full-catalog-discovery-on-google) - thredUP increases sell-through by 15% with enriched product data (https://www.lily.ai/case-studies/thredup-sell-through-15-percent) - A global luxury retailer increases relevant results for descriptive searches by up to 30x (https://www.lily.ai/case-studies/luxury-retailer-descriptive-search-30x) - A large multi-brand retailer boosts product discovery, projecting a $22M revenue increase (https://www.lily.ai/case-studies/multi-brand-retailer-22m-product-discovery) - Lily AI recommendations increase RPV by 0.65%, driving $20-30M in incremental demand (https://www.lily.ai/case-studies/multi-brand-retailer-rpv-recommendations) - Lily AI demand forecasting drives 7-8 digit revenue lift with accuracy at scale (https://www.lily.ai/case-studies/demand-forecasting-accuracy-at-scale) - How a luxury home brand boosted sales from Shopify site search and Google search (https://www.lily.ai/case-studies/luxury-home-brand-shopify-and-google) - Tapestry boosts traffic and sales from Google Search with Lily AI (https://www.lily.ai/case-studies/tapestry-google-search-discovery) - From experimentation to full AI vibes at Rue Gilt Groupe (https://www.lily.ai/case-studies/rue-gilt-groupe-full-ai-vibes) - From SEM to GEO: how Merrell connects with Gen Z and millennials (https://www.lily.ai/case-studies/merrell-sem-to-geo) - How Bloomingdale's makes site search convert with enriched product data (https://www.lily.ai/case-studies/bloomingdales-site-search-convert) - “AI is here, now what?” with Tapestry and Lily AI (https://www.lily.ai/case-studies/tapestry-ai-is-here-now-what) - How J.Crew uses AI to delight customers and power sales growth (https://www.lily.ai/case-studies/jcrew-ai-delight-customers) --- # Agentic commerce: what it is, what changed in 2026, and what a brand actually has to do https://www.lily.ai/blog/agentic-commerce 2026-08-25 Agentic commerce is AI agents buying on a shopper's behalf. What changed in 2026, what an agent reads from your catalog, and what to fix before Q4. Agentic commerce is online purchasing in which an AI agent acts on the shopper's behalf, reading structured product data to discover, compare and in some cases buy a product without the shopper ever browsing a store. The shopper states an outcome, not a query, and the agent resolves it against whatever product records it can actually parse. That last clause is the whole commercial question, because an agent cannot shortlist a product whose record does not answer the constraint it was given. Most writing on this subject is about what agents will do to retail. This page is about what a brand has to change, and by when. ## What is agentic commerce, in plain terms? Agentic commerce is the model where a shopper delegates the work of shopping to software. Instead of typing "black linen dress" into a search box and filtering the results, the shopper says something closer to find me an unlined black midi dress for a July wedding, under $300, that arrives by Friday, and an agent goes and resolves it. The shift that matters is not conversational phrasing. It is that the constraints move from the filter sidebar into the request itself. "Unlined," "for a July wedding," "arrives by Friday" are all conditions the agent has to check against something. If your record carries fabric, lining, occasion and dispatch data, you are checkable. If it carries a style code and a house color name, you are not. Agentic AI commerce and agentic commerce are used interchangeably. The distinction some writers draw between them is not load-bearing. ## How is agentic commerce different from traditional e-commerce? The difference is who reads your product record, and what they do when it is incomplete. A shopper who lands on a thin product page can still buy, because a photograph and a price are enough for a person. An agent working from the same record has nothing to evaluate and moves on silently. The same catalog, read two ways. The failure mode in the right-hand column produces no error message. | Traditional e-commerce | Agentic commerce Who initiates | Shopper browses or searches | Shopper states an outcome; agent executes Input | Keywords, then filters | A constrained natural-language request What decides visibility | Page relevance, bid, merchandising | Whether the record can satisfy the stated constraints What a thin record costs | Some friction; the shopper compensates | Silent exclusion from the shortlist Where the brand intervenes | On-site merchandising, creative, bid | The product-language record, upstream of all of it Feedback when you fail | Bounce rate, low conversion | Nothing. You were never in the set That last row is the reason this is difficult to manage. Every other channel tells you when it is going badly. Agentic surfaces do not produce a disapproval, a warning, or a drop you can attribute. The product simply stops being an option, and the dashboard looks the same as it did last week. ## What actually happened in agentic commerce in 2026? The infrastructure arrived faster than the demand did, which is why the market feels simultaneously overhyped and urgent. Here is the record, with dates. The agentic commerce timeline, September 2025 to August 2026. Kept current; see the changelog at the foot of this page. When | What shipped | What it means for a brand Sep 2025 | OpenAI launches Instant Checkout and the Agentic Commerce Protocol (ACP) with Stripe | The first serious buy-in-chat rail Nov 2025 | Walmart puts roughly 200,000 products through Instant Checkout | Enterprise retail tests the pipe at scale Jan 11 2026 | Google announces the Universal Commerce Protocol (UCP) with 20+ backers, the Business Agent in Search, and Direct Offers in AI Mode | Google formalizes the agent-readable product record Feb 2026 | Forrester finds only about 30 Shopify merchants actually live on Instant Checkout | The gap between announced and real Mar 2026 | OpenAI steps back from in-chat checkout, moving toward Apps | The consensus forms: discover in AI, buy on site May 7 2026 | Google retires FAQ rich results from Search | Structured Q&A still matters for machines, not for SERP stars May 19 2026 | Google I/O: Universal Cart across Search, Gemini, YouTube and Gmail, plus Agent Payments Protocol (AP2) updates | One cart, many surfaces May 20 2026 | Google Marketing Live: six conversational attributes in Merchant Center, and AI Performance Insights, which reports a brand's share of voice on AI surfaces | The product record becomes both an input and a measurable one May 2026 | Amazon folds Rufus into Alexa for Shopping with agentic purchase | A second major agent reading catalogs Aug 2026 | UCP moves to a published quarterly release cadence, with fulfillment and payment work targeted ahead of the holiday season | Q4 is the deadline the ecosystem is working to Two things are worth reading out of that table rather than skimming past. First, every one of those releases is a pipe, and not one of them supplies the quality of what flows through it. Google built the protocol, the cart, the payment rail and the reporting. It did not write your product descriptions. The input is the operator's job and it has stayed the operator's job through the entire build-out. Second, May 2026 is the inflection, not January. January gave the market a protocol. May gave it a set of fields to fill in and a report that shows whether filling them in worked. The budget follows measurement, and measurement arrived in May. ## What does an agent actually read when it picks a product? An agent reads two layers: the feed you send out and the page it crawls. It resolves the shopper's constraints against structured fields first, because structured fields are cheap to check and prose is expensive to interpret. In practice, that means the following, in roughly this order of influence: 1. Title. The first thing scored against the request. Internal style (style numbers, house color names, brand-first construction) costs you here more than anywhere else. 2. Category and product type. Where the item sits in the commerce graph. A top-level bucket gets you benchmarked against the wrong products. 3. Filterable attributes. material, pattern, size_system, age_group, gender, color. These are the fields a constraint resolves against. 4. The richer structured fields. product_highlight, product_detail, and now the conversational attributes. Most catalogs ship these blank. 5. The crawlable page copy. Read for what the feed does not carry, and check for agreement with the feed. The six conversational attributes Google added in May 2026 (question_and_answer, document_link, related_product, item_group_title, variant_option, popularity_rank) sit squarely in layer four. They are optional, they do not affect product approval, and most merchants have not filled them in. ## What goes wrong when a product record is thin? Two failure modes, and the second is worse than the first because it is loud. The quiet failure is exclusion. The product is fine, the price is fine, and it is simply never a candidate, because nothing in the record answers the constraint the shopper stated. There is no signal. You find out at the end of the quarter, if at all. The loud failure is misreading. An agent working from an ambiguous or incomplete record does not abstain, it infers. Where a person reading a confusing spec or a partial ingredient list asks a question or moves on, an agent produces an answer with confidence and acts on it. That inference reaches the shopper as a statement of fact about your product, and you did not write it. It is the worse of the two failures, because exclusion is silent and this one is wrong out loud. Our position, stated plainly: every company selling services in this area is working it out as it goes, because the surfaces are new and nobody has the full picture of how citation and selection actually resolve. What is knowable is which inputs those systems read, and whether yours are complete. Start there rather than with a theory of the algorithm. ## Which agentic commerce platforms and protocols matter? Four protocols carry most of the traffic in this conversation, and they do different jobs. Confusing them is the most common error in vendor material. The four protocols, what each one governs, and who backs it. Protocol | What it governs | Origin ACP (Agentic Commerce Protocol) | Product discovery and merchant handoff in chat | OpenAI with Stripe, Sep 2025 UCP (Universal Commerce Protocol) | Agent interaction with merchant catalogs and checkout | Google, Jan 2026, 20+ backers AP2 (Agent Payments Protocol) | Payment authorization by an agent | Google, updated May 2026 MCP (Model Context Protocol) | How models connect to external data sources and tools | Open standard, widely adopted None of these is a strategy. They are plumbing, and they are converging: the ecosystem has moved to a published release cadence with fulfillment and payment work aimed at the holiday season. What a brand controls is not which protocol wins. It is whether the record travelling down all of them is complete. ## What should a brand fix before Q4? Work this in order. The first three cost nothing but time and gate everything after them. 1. Audit titles against how people actually ask. Pull your top 200 SKUs by revenue and read the titles cold. If a title needs internal knowledge to decode, rewrite it so the most-searched descriptor comes first. 2. Fill the filterable attributes. material, pattern, size_system, age_group, gender. These are the fields constraints resolve against and they are empty in most catalogs. 3. Deepen google_product_category and product_type. A two-level internal path gives bidding nothing to act on and puts you in the wrong benchmark set. 4. Populate the six [conversational attributes](/blog/merchant-center-conversational-attributes). Optional, zero risk to approval, and largely unclaimed. Mine your own support tickets and PDP Q&A for the question_and_answer values: the answers your team already types every week are the source material. 5. Check the page and the feed. An agent reads both. Where they disagree, you look unreliable. 6. Open AI Performance Insights in the Merchant Center and read your share of voice against similar brands before you change anything, so you have a baseline. 7. Re-run all of it on a cadence. A catalog is a population, not a project. New SKUs arrive with the original defaults, and any pass you completed in June is already decaying. Step seven is the one that turns a successful project into a flat year. ## How do you prove agentic work actually worked? Against a control, or not at all. This is the part of agentic commerce most likely to be sold on a score rather than a result, and a score is not an outcome. The workable method is unglamorous: hold spend constant, change only the product-language record for a treatment group, and compare against an untouched holdout over a named window. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control. On Meta Advantage+, a holdout test cross-validated against Meta's own Conversion Lift study measured a 21.4% improvement in ROAS. Those are paid-surface results, stated as such deliberately. On agentic and answer-engine surfaces the honest claim is narrower: you can improve readiness and input quality, and you can measure whether your share of voice moved. Placement, citation and final selection sit with the platform. Any vendor telling you otherwise is selling a guarantee they do not control. ## Would it survive a finance review? That is the question to carry into every agentic commerce conversation this year, and it disqualifies most of what is currently being sold. A CFO asked to fund agentic readiness will ask four things, and they are the same four whether the surface is Google Shopping or an assistant nobody had heard of last spring. Are you fixing the input, or a layer above it. Does the fix run continuously, because the catalog churns. Does it cover the surfaces that pay this quarter, not only the interesting ones. And can you show lift against a control. An agent cannot recommend a product it cannot parse, and it will not tell you when it skipped you. The work is to make the record answerable, then to prove the answer moved something. Start with your top 200 SKUs and the five empty attribute fields, and get a baseline out of AI Performance Insights before you touch anything. ## FAQ **What is agentic commerce?** Agentic commerce is online purchasing where an AI agent acts on a shopper's behalf, reading structured product data to discover, compare and sometimes buy a product. The shopper states an outcome rather than a search term, and the agent resolves it against whatever product records it can parse. **What is the difference between agentic commerce and traditional e-commerce?** In traditional e-commerce a shopper browses and filters, so a thin product record creates friction the shopper can work around. In agentic commerce the agent checks stated constraints against structured fields, so a thin record results in silent exclusion from the shortlist with no error and no signal. **Is agentic commerce actually happening yet, or is it hype?** Both are true at once: the infrastructure shipped fast across 2025 and 2026, while live merchant adoption has lagged the announcements considerably. The practical position is that discovery through AI is real now, agent checkout is still early, and the product-data work required is identical either way. **What is agentic commerce optimization?** Agentic commerce optimization is the practice of preparing product data, feeds and page content so AI agents can understand, evaluate and recommend a catalog. It is an extension of the same product-language work that drives paid and onsite performance, not a separate discipline with separate inputs. **Which protocols matter in agentic commerce?** ACP governs discovery and merchant handoff in chat, UCP governs how agents interact with merchant catalogs and checkout, AP2 governs agent payment authorization, and MCP governs how models connect to external data. None of them determines the quality of the product record travelling through them, which stays the merchant's responsibility. **How do I get my catalog ready for shopping agents?** Start by auditing titles on your top revenue SKUs, filling the empty filterable attributes, and deepening your category and product type paths. Then populate the six Merchant Center conversational attributes, confirm your feed and product page agree, and take a share-of-voice baseline before making further changes. --- # Does feed optimization increase sales? The anatomy of a controlled test https://www.lily.ai/blog/does-feed-optimization-increase-sales 2026-08-25 Does feed optimization increase sales? Yes, against a control. Get the full test design: matched spend, 28-day holdout, difference-in-differences. Does feed optimization increase sales? Here is the controlled answer, method first. Yes, but only when measured against a control. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue versus the untouched control group; a Meta Advantage+ holdout test cross-validated with Meta's own Conversion Lift returned +21.4% ROAS. Platform-reported lift is not evidence: it attributes to itself. The only trustworthy answer to "did feed optimization increase sales" comes from a holdout, a matched spend, and a difference-in-differences read. What follows is the whole method: how to build the holdout, what to freeze, how to read the result, and what it does not prove. ## Why is this question so hard to answer honestly? Because almost every number you will be shown was produced by a party with an interest in the outcome, and because ecommerce revenue moves for a dozen reasons at once. A feed change ships on a Tuesday, revenue is up by Friday, and nothing tells you whether that was the feed, a rival's stockout, a promo email, or a platform update. Attribution has gotten harder, not easier. Discovery spans four surfaces that all read the same product-language record (Google Shopping, Meta Advantage+, onsite search, and AI assistants) and shoppers cross between them before buying. Semrush found that 22% of US shoppers have bought inside an AI tool, while 50% bought elsewhere after researching one. That half never appears in your analytics. That leaves two options: accept the platform's own report of how well the platform did, or build a control group. Incrementality testing in ecommerce exists because the first option is a conflict of interest wearing a dashboard. ## What does a valid feed optimization A/B test look like? A valid test changes exactly one thing, in one half of a matched population, over a fixed window, with spend held constant. Everything else is a case study. Here is how to design one, so you can run it yourself or demand it from a vendor. ### Step 1: Build the matched population Stratify the catalog before you split it. Group products by whatever predict revenue independently of language quality (category, price band, conversion rate, inventory depth and trailing revenue are the usual candidates) then split within each stratum so both arms have the same shape. Random splitting across a whole catalog is the most common way a feed optimization A/B test breaks. Ecommerce revenue concentrates on a few products, so an unstratified coin flip can hand one arm a disproportionate share of the top sellers and decide the outcome in advance. ### Step 2: Freeze everything that is not the input The treatment arm gets the rewritten product-language record: titles, attribute values, product type assignment, description language. The control arm gets nothing. Read this table left to right: only the first column should change; everything in the second stays frozen so it cannot explain the result. Changes in the treatment arm | Held constant in both arms Titles, rewritten in shopper vocabulary | Bids, budgets, bid strategy Attribute values and coverage | Campaign structure, targeting, geography Product type and taxonomy assignment | Landing pages, pricing, promotions Description language | Creative, assets, feed rules Nothing else | Shipping settings, inventory availability Matched spend is the part people skip. If the treatment arm gets more budget because it started performing, you have measured budget, not language. Fix spend at campaign level for the duration; a mid-test change voids the read. ### Step 3: Run a fixed window and pre-commit to it Set the window before the test starts and do not move it. Twenty-eight days is a defensible floor: the window has to cover at least one full purchase cycle for the category and absorb the lag between a re-crawled listing and a stabilized impression profile. A test that runs until the number looks good is not a test, it is a search for a favourable stopping point. ### Step 4: Read it as a difference-in-differences Do not compare the treatment arm after against the treatment arm before. Compare the change in the treatment arm to the change in the control arm over the same window. The arithmetic is deliberately boring: the incremental effect is the treatment arm's delta minus the control arm's delta. If the whole category rose because it was August, the control rose with it and that movement cancels out. That subtraction is the entire reason a holdout exists. ## What did the controlled tests return? In a matched-spend A/B test with a 28-day holdout, the rewritten product-language input produced a 28% increase in Google Shopping revenue against the untouched control. On Meta Advantage+, a holdout test cross-validated against Meta's own Conversion Lift measurement returned a 21.4% improvement in ROAS. On the retailer's own site, a statistically significant A/B test returned a 28.3% increase in onsite revenue. Three surfaces, three separate control groups, three separate reads. Lily has run more than 1,000 controlled tests of this design before publishing the benchmark, which is why these figures describe a range of outcomes rather than one hero case. The Meta figure is worth pausing on because it was cross-validated: the independent holdout and Meta's own Conversion Lift methodology agreed. That agreement is worth more than either read alone, and it is a check you can ask any vendor to run. ## What does a confidence interval actually tell you? A confidence interval tells you the range of true effects consistent with what you observed; the headline percentage is only the middle of that range. A point estimate with no interval is the midpoint of a distribution the vendor is not showing you. Two vendors can report the same headline lift while one has an interval sitting comfortably above zero across its whole span and the other has a lower bound near nothing. The point estimate is identical; the business case is not, because you budget against the lower bound. So ask three things: how many converting products were in each arm, how wide the interval is, and where its lower bound sits. A wide interval is not dishonest, it is what a short or small test honestly produces. A missing interval is. ## What this result does not prove It does not prove that rewriting your catalog will produce 28%. It proves that in these catalogs, on this surface, over these windows, changing only the product-language input moved revenue against a control that did not get it. : It is not a universal effect size. Lift depends on how far the starting catalog sits from shopper vocabulary; a catalog already written in customer language has less headroom. : It does not decompose the mechanism. Titles, attribute coverage and taxonomy moved together by design, so the test cannot say how much each contributed. : It does not transfer across categories. Apparel, beauty and home have different query behavior and attribute vocabularies, so a result in one is a hypothesis in another. : It does not prove a permanent effect. Effects measured over 28 days are evidence about 28 days. : It says nothing about margin. A test that holds spend constant answers the revenue question only. Publishing that list costs nothing and is the point. A claim you can bound is a claim a finance team can use. ## How do you calculate product feed optimization ROI? Divide the incremental revenue from the difference-in-differences read (not the total revenue in the treated group) by the fully loaded cost of the work, including internal hours. The most common error in product data ROI maths is putting the treated group's whole revenue in the numerator, which credits the change with sales that would have happened anyway. Then apply contribution margin, not gross revenue, if the number is going to a CFO. Incremental revenue at category margin, minus the cost of the work, over a stated window, with a control group named: a finance review can check that sentence. ## How to measure feed optimization when someone else ran the test Ask for the design, not the headline. Read this table left to right: the shape of the claim you will hear, and the one thing that would let you check it. What the claim sounds like | What would make it checkable "Customers see up to X% lift." | How many accounts, and the median account's result rather than the best one. "Revenue grew X% after we launched." | A control group that did not get the change over the same window. "The platform reported an X% conversion increase." | An independent holdout, because the platform is scoring a campaign it delivered. "The result was statistically significant." | Significant against what null, at what sample size, and the interval's width and lower bound. "ROAS improved by X%." | Whether spend was held constant, and whether campaign mix and geography moved. "It worked for a retailer like you." | Catalog size, category, and whether the window overlapped a promotion. "Results within X days." | The stopping rule, pre-committed in writing before the test began. If a vendor cannot fill the right-hand column for their own headline, you have learned something useful. Run it on us too. ## Why the answer decays if you stop A single controlled result measures a catalog at a moment, and catalogs do not hold still. New SKUs land, assortments rotate, suppliers rewrite their descriptions, and the surfaces change what they read: Google announced the Universal Commerce Protocol in January 2026 and the six Merchant Center conversational attributes in May 2026, a new set of fields for the same catalog to answer. So the honest framing is a maintained input, not a project with an end date. A one-time rewrite has a half-life: the population it was measured on is replaced product by product until the measurement describes a catalog that no longer exists. That is why continuous operation is a buying criterion rather than a feature, and why product data enrichment matters more than any single rewrite. ## So does feed optimization increase sales, or not? Yes, when it changes the language layer rather than the formatting layer, and yes, when it is measured against a holdout instead of reported by the platform that served the ads. Everything above is the apparatus for telling those two cases apart. The next step is small: pick one surface, stratify one catalog, freeze the spend, hold back a control, pre-commit to 28 days. Then ask every vendor in your evaluation, us included, one question: would this survive a finance review? ## FAQ **Does feed optimization increase sales?** Yes, when the change is measured against a holdout rather than reported by the platform that delivered the ads. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control group. **How do you measure product feed optimization ROI?** Split the catalog into matched treatment and control groups, hold spend and campaign settings constant, and read the difference between the two groups' changes across a fixed window. Divide the incremental revenue from that read, not the total revenue in the treated group, by the fully loaded cost of the work. **What is a holdout in a feed optimization A/B test?** A holdout is a group of products deliberately left unchanged so it can show what would have happened without the optimization. Without one, any revenue movement is confounded with seasonality, promotions, price changes and platform algorithm updates. **Is platform-reported lift enough evidence for a finance review?** No, because the platform reporting the lift is scoring a campaign it also delivered, which makes it an interested party rather than an independent measurement. A finance review needs an incremental read from a control group the platform did not touch. **Is feed optimization worth it for a small catalog?** It can be, but a small catalog makes the test harder, because fewer converting products means a wider interval around the estimate. Size the test on the number of converting products rather than total SKUs, and extend the window instead of accepting a noisy read. **How long should a feed optimization test run?** Long enough to cover at least one full purchase cycle for the category and to let re-crawled listings stabilize, which is why the Google Shopping test used a 28-day window. Shorter windows tend to measure indexing lag rather than shopper response. --- # How AI Shopping Assistants Actually Pick Which Products to Recommend https://www.lily.ai/blog/how-ai-shopping-assistants-pick-products 2026-08-25 AI shopping assistants retrieve before they rank. See what each surface reads from your product record, and which missing attributes disqualify you. AI shopping assistants do not browse; they retrieve. When a shopper asks for a two-person three-season backpacking tent under 2.5 kg that pitches freestanding, for a wet October in Scotland, under $400, an assistant matches that description against structured product records it has indexed (feeds, merchant data, retailer catalogs, crawled pages) and can only recommend products whose records answer the question. A product that never states packed weight, pitch type or wet-weather performance is not ranked lower; it is not a candidate. Visibility in AI is a symptom; the product-language record is the lever. That is the mechanism. The rest of this page is the detail behind it: what the retrieval chain actually does, what each surface is known to read, and (the section most articles on this subject skip) which parts of the system are genuinely opaque and cannot be controlled by anyone, including us. ## How do AI shopping assistants go from a question to a recommendation? The assistant decomposes the question into constraints, retrieves candidate products whose structured records can satisfy those constraints, and only then compares the survivors. Two of those three steps happen before your product is ever weighed against a rival's. Take the query above and break it the way a retriever has to break it. "A two-person three-season backpacking tent under 2.5 kg that pitches freestanding, for a wet October in Scotland, under $400" is not one request. It is eight separate constraints, each of which has to be checked against something written down. Now take the product record most outdoor catalogs actually ship. Title: 2-Person Dome Tent - Alpine Green. Category: Sporting Goods > Outdoor Recreation > Camping & Hiking > Tents. Product type: Tents > Dome. Color: alpine green. Capacity: 2 person. Price: $379. Availability: in stock. (A constructed example, not a real SKU, but nothing has been left out of it to make the point.) That is a complete, healthy, non-erroring record. It will pass every feed diagnostic you run. Here is that same record measured against the question the shopper actually asked. One shopper question, broken into the eight constraints a retriever has to satisfy, and whether a typical tent record can answer each one. Constraint in the question | The field that would answer it | Present in the record? | What the retriever does Tent | google_product_category, product_type | Yes | Passes Two-person capacity | capacity, title | Yes | Passes Under $400 | price | Yes | Passes ($379) Three-season rating | product_type: "Tents > Dome" | Ambiguous | A dome is a pole geometry, not a season rating; no rating is stated anywhere Packed weight under 2.5 kg | No field; absent from the description | No | Cannot be evaluated Freestanding pitch | No field; no pitch-type guidance anywhere | No | Cannot be evaluated Wet-weather performance | No field for waterproof rating or fly coverage | No | Cannot be evaluated October / shoulder-season use | No field | No | Cannot be evaluated Four of eight, and one of the four is ambiguous. The physical tent behind a record like that is in stock, priced correctly, and would satisfy this shopper completely if they could pick it up and read the spec panel on the box. The assistant has no way to know that, because the record never says it. ## Is it ranking, or is it retrieval? Retrieval decides whether you are a candidate at all; ranking decides the order of the candidates. Nearly everything written about brand visibility in AI addresses the second problem, and nearly every product that loses, loses at the first. This is the part that breaks a decade of search intuition. In classic organic search, a thin page is still on the list, just lower down, and the remedy is to grind upward. Retrieval against structured records does not work that way. If the query carries a constraint your record cannot answer, you are not a weaker match: you are outside the set being compared. An absent attribute is not a neutral omission. It is a silent disqualification from the consideration set. Silent is the operative word. There is no rejection notice, no impression log, no diagnostic in any console that says excluded: occasion not stated. Feed health tooling will report the record as perfect, because by the standards of feed health it is perfect. The gap only becomes visible when someone sits down with a real query and reads the record against it. The same logic runs one level up, at the category. If your hierarchy files a backpacking tent under a buying-team label rather than something a shopper would say, the retrieval step can miss you before any attribute is even consulted. That is a product taxonomy problem disguised as an attribute problem. ## What each surface reads Every assistant assembles its answer from some combination of three inputs: a merchant-supplied product feed, the crawled product page and its structured markup, and third-party sources it can cite. The mix differs by surface, and the mix decides which of your fixes reaches which assistant. The same product record, read four different ways, plus the paid surfaces reading it this quarter. Only the last column contains a lever. Surface | What it reads | What it does with it | What an operator controls Google AI Mode and the Business Agent | Merchant Center product data, including the conversational attributes added on 20 May 2026; the crawled product page and its Product markup; the Shopping Graph | Answers the shopper inside the results surface, serves Direct Offers in AI Mode, and (through the Business Agent) holds a brand-voice sales conversation on the retailer's behalf | Feed coverage and accuracy, population of the conversational attributes, page markup, category mapping ChatGPT and ACP | Crawled product and category pages, structured markup, merchant-supplied product data where a merchant participates, and third-party editorial and review sources | Composes a recommendation set inside the conversation. The Agentic Commerce Protocol, launched with Stripe in September 2025, is the checkout rail, not the retrieval mechanism | What the page and the record state, whether the page is crawlable, whether an independent source can verify the claim Perplexity | The live web, restricted to pages it can cite: retailer product pages, editorial roundups, review sites | Answers with sources displayed, which means a citable third party can outweigh your own page in the answer | The specificity of your own page, plus what a reviewer can factually state because your record stated it first Amazon's Alexa for Shopping | Amazon's own item catalog: submitted attributes, bullets, browse-node placement, enhanced content, customer reviews and Q&A | Answers in-app and completes an agentic purchase inside Amazon, following the May 2026 folding of Rufus into Alexa for Shopping | The completeness of the item record you submit to Amazon, and nothing outside it Google Shopping, Performance Max, Meta Advantage+ (paid, same record) | The same Merchant Center or commerce catalog feed | Machine-matches the record to auctions, queries and audiences | The same fields. This is why the work pays before the assistant referrals arrive Two honest notes on that table. First, the entries describe what each system is known to consume, drawn from the platforms' own announcements and documentation, not from any privileged access. Second, none of these companies publishes its retrieval function, its weighting, or its candidate-set size, and none exposes an impression log. Anyone presenting a ranking-factor list for these surfaces is describing a black box from the outside and calling it a schematic. ### Where the systems are genuinely opaque You control what your record says. You do not control what happens after it is read. The split is worth writing down, because the second column is where most budgets go to die. In your control: whether an attribute is stated at all; the vocabulary it is stated in; which structured fields are populated and how completely; the markup on the page; feed freshness and coverage across the whole catalog rather than the top sellers; and what a third party can truthfully say about the product, because you gave them the material. Not in anyone's control: the retrieval function and its weighting; candidate-set size; whether an answer cites you, paraphrases you, or links to a roundup that mentions you; personalization by user, session and location; which model version answered; and whether the same prompt returns the same set tomorrow. It generally will not. That second list is not a caveat we are adding for modesty. It is the reason the first list is the only place to spend. ## Why do "AI visibility" dashboards not move the number? A visibility dashboard samples prompts and reports whether you were mentioned. That is a thermometer: genuinely useful for noticing that something changed, structurally incapable of causing the change. The measurement is also weaker than it looks. Assistant answers vary by user, session and model version, so a sampled mention rate is an estimate of an estimate. Worse, mention is not the outcome anyone is buying. A brand can be named in an answer and still not have a single product retrieved into the recommendation set, because those two things are decided by different mechanisms: one by what the web says about you, the other by what your product record can answer. This is where answer engine optimization for ecommerce diverges from its content-marketing cousin. The content version is about being cited as a source. The commerce version is about being retrievable as a product, and no amount of authority substitutes for a packed-weight field that does not exist. If a dashboard number moves and you cannot say which change moved it, you have a reading, not a result. Our position on that is the same everywhere: a change is only real when it is measured against a control. ## Why fix this now, when AI referrals are still a rounding error? Because the record you fix for the assistants is the same record the paid surfaces read this quarter, and because the discovery behavior has already moved even though the checkout behavior has not. AI-referred traffic is still a small fraction of retail sessions. That is an argument about timing, not about direction. In March 2026, OpenAI stepped back from in-chat checkout and moved toward Apps, and the consensus that formed across the industry was discovered in AI, buy on site. That is the load-bearing point on this page. The assistant is turning into the place where the consideration set is built, while the transaction stays on the retailer's own property. Semrush's data points the same way: 50% of US shoppers bought elsewhere after researching inside an AI tool. That is the behavior you would expect from a surface that assembles the consideration set and then hands the transaction back to the retailer, and from traffic that arrives having already been filtered against eight constraints. Three things follow, and they are why this is a rational spend now rather than a 2028 bet. 1. The input work compounds. An attribute written today is read by every surface that arrives afterwards. Merchant Center conversational attributes did not exist before May 2026, and the retailers who could populate them in week one were the ones whose records already carried occasion and fit language for other reasons. 2. The same record pays this quarter. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control. In a statistically significant A/B test, the same class of change produced a 28.3% increase in onsite revenue. We do not publish an equivalent controlled figure for assistant-referred revenue, because the surfaces do not yet expose the attribution needed to build one honestly. When they do, we will show the method first. 3. It decays. Catalogs churn: new SKUs land weekly, seasonal assortments rotate, suppliers rewrite their own descriptions, and the surfaces keep adding fields. A one-time enrichment project is a snapshot of a catalog that has already changed underneath it, which is why this has to run continuously rather than as a Q3 initiative. ## How do you audit your own product record this week? Run it by hand on twenty SKUs before you buy anything. The exercise takes an afternoon and it is the only diagnostic that surfaces the failure described on this page, because every automated feed check will pass a record that is silently unretrievable. 1. Pull your twenty highest-revenue SKUs and export the full record for each: title, description, category, product type, every populated attribute. 2. For each SKU, write the five questions a real shopper would say out loud. Not keywords. Spoken sentences, with constraints in them. 3. Mark every constraint in every question as present, absent, or implied in the record. 4. Apply the scoring rule that makes this honest: implied counts as absent. A retriever cannot infer that a dome tent pitches freestanding, that a listed weight is the packed weight rather than the trail weight, or that a fly reaches the ground in wind. If a human has to reason from the photo, the constraint is absent. 5. Count the absences. That count is a measure of how many of the shopper's stated constraints your record cannot answer, and each absent constraint is a query the record cannot be matched against. The five questions that generalize across most catalogs: what occasion or use is this for; how does it fit or size; what is it made of and how is it constructed; what are the care and durability constraints; and what does it work with, fit into, or sit beside. If your export cannot answer those, the fix is in the record, not in the campaign. From there the work is ordinary and unglamorous: get the attribute set right, get it delivered, populate the fields the newest surface asked for, and keep the whole library in one place. The platforms built every pipe: UCP, ACP, Universal Cart, conversational attributes, AI Mode, Alexa for Shopping. Not one of them supplies the quality of what flows through. That is the operator's lever, and it is the only part of this system that answers to a budget. Judge any spend against it on four things: does it fix the input rather than the layer above it, does it run continuously because catalogs churn, does it cover the surfaces that pay this quarter as well as the interesting ones, and does it prove lift against a control. Or, compressed into the question a CFO will ask anyway: would it survive a finance review? ## FAQ **How do AI shopping assistants decide which products to recommend?** AI shopping assistants break a shopper's question into constraints, retrieve products whose structured records can satisfy those constraints, and then compare only the products that survived retrieval. A product whose record does not state an attribute the question depends on is not ranked lower; it is excluded from the comparison entirely. **Does ChatGPT read my product feed?** ChatGPT composes product recommendations from crawled product pages, structured markup, third-party editorial and review sources, and merchant-supplied product data where a merchant participates in a shopping integration. The Agentic Commerce Protocol that OpenAI launched with Stripe in September 2025 governs checkout rather than retrieval, so participating in it does not by itself make a product findable. **How do I get my products recommended by AI?** State every attribute a shopper would name in the question inside the product record itself, using the words the shopper would use rather than internal merchandising vocabulary. Then keep doing it, because catalogs turn over continuously and the surfaces keep adding fields, such as the Merchant Center conversational attributes that did not exist before May 2026. **Is optimizing for AI shopping assistants the same as SEO?** Search engine optimization mostly improves position among pages that already qualify, whereas AI shopping assistants decide qualification first, at the level of the product record. Tactics aimed at the ranking layer cannot repair a record that never stated the attribute the shopper asked about. **Can I track whether my products appear in AI shopping assistants?** Sampling tools can report whether a brand or product was mentioned across a set of prompts, which is a useful change detector but not an attribution method. None of the major assistants publishes impression, candidate-set or ranking data, so the only claim that survives scrutiny is a change measured against a control on a surface that does report. **Is this worth doing while AI-referred traffic is still small?** The same product record that AI shopping assistants read is the record Google Shopping, Performance Max and Meta Advantage+ read today, so the work pays on the paid surfaces in the current quarter regardless of assistant volume. Waiting also forfeits compounding, because a record enriched now keeps earning as each new surface arrives, while a record fixed in 2028 starts from zero in 2028. --- # What actually goes in Google Merchant Center conversational attributes https://www.lily.ai/blog/merchant-center-conversational-attributes 2026-08-25 What actually goes in Google Merchant Center conversational attributes: worked before-and-after values for apparel and home, plus where to mine them. [image] Conversational attributes are Merchant Center fields Google introduced in 2026 that let a product record answer the questions a shopper would ask a salesperson (what it works with, what to buy instead, what a product question's answer is) rather than only stating its specifications. They exist because AI Mode and the Business Agent need a product record that can hold a conversation, and the standard attribute set was written for a keyword-matching era. Populating them well requires shopper-question language, not catalog language. Google Merchant Center conversational attributes are new, and most coverage stops at the announcement. This page is about the values. ## When did Google add conversational attributes, and why? Google shipped this in two waves. On 11 January 2026 it announced the Universal Commerce Protocol (UCP), the Business Agent (a brand-voice sales associate inside Search) and Direct Offers in AI Mode: the protocol and the surface. At Google Marketing Live on 20 May 2026 it shipped the fields those surfaces read, six new Merchant Center conversational attributes, rolling out globally. Together they let a shopper have a sales conversation with your catalog instead of scrolling a grid. The Business Agent has to say something when a shopper asks whether the chair fits under a 29-inch counter, and it can only say what the record says. Google built the surface, the protocol and the fields, not the sentences: the gap under every surface a catalog feeds. ## How is a conversational attribute different from the product description? A description is one block of prose for a human already on your page, read whole. A conversational attribute is a discrete, question-scoped value a machine can retrieve and quote without summarizing a paragraph and guessing right. The difference is mechanical, not stylistic. Answering "is this rug safe over underfloor heating?" from a description requires inference, and inference is where invented product claims come from: a returns problem before a marketing one. Conversational attributes are the opposite trade: narrow, literal, quotable. They do not replace the standard attribute set, covered field by field in product attributes. ## Which shopper questions do Google Merchant Center conversational attributes answer? Every conversational attribute answers a question type. What makes a value quotable is a number, a condition or a named limitation; marketing adjectives answer nothing. The question a shopper asks, the question type it belongs to, and what separates a quotable value from an unusable one. The question types below are our grouping, not a published Google taxonomy. Shopper question | Question type | Weak value | Value that can be quoted "Will this wrinkle in a suitcase?" | Answer to a product question | "Premium linen blend" | "Wrinkles readily; creases drop out in 20 minutes with steam." "What do I need to buy with it?" | Compatible accessories | (empty) | "Requires a 1/4-inch felt rug pad on hard floors." "Is something else better for me?" | Substitutes | (empty) | "For homes with a robot vacuum, the flat-weave version." One more question belongs under the table rather than in it, because it is not a question type at all. "What's the catch?" is a posture, and most records leave it unwritten. The quotable answer reads like "sheds for four to six weeks; vacuum without a beater bar." Omitting it is the default choice and the expensive one. ## What does a good record look like before and after? Two categories where the gap is widest. Before is what a well-run catalog looks like today. ### Apparel: a linen midi dress Before. Color, size run, material, care, and 140 words about effortless summer dressing. Feed status green. After, in the conversational layer. : Answer to a product question: "Unlined. Sheer in direct sunlight; most customers wear a slip." : Occasion suitability: "Daytime outdoor events and garden weddings. Not warm below 15°C." : Substitutes: "For cooler months, the heavier washed-linen weave." ### Home: an 8x10 hand-loomed wool rug Before. Dimensions, pile height, 100% wool, color family, weave type, 160 words about artisanal craft. After, in the conversational layer. : Answer to a product question: "Sheds for four to six weeks; vacuum weekly, never on a beater-bar setting." : Compatibility: "Safe over underfloor heating to the floor manufacturer's limit. Use a low-profile felt pad, not rubber-backed." : Use-case suitability: "Dining rooms: chairs slide with some resistance at this pile height." That last line costs a few dining-room sales. It also makes your record the one the assistant trusts on the next question. ## Where do you find the answers Google now wants? You already have them, written by customers, in four places nobody in merchandising reads. Mining them is a five-step job, not a content project. 1. Twelve months of pre-purchase support tickets and chat transcripts. Group by category, not SKU; a small number of questions per category cover most of the volume. 2. Free-text return reasons. Coded reasons ("wrong size") are useless. The typed box is where "I didn't realize it was unlined" lives, and each one is a missing attribute value. 3. Reviews and product-page Q&A filtered to a question mark or "wish I'd known." That filter surfaces the honest-caveat values. 4. Onsite searches that returned results but got no clicks. Questions your catalog answered badly rather than missed. 5. Write once at category level, vary at SKU level. Wool rugs shed: written once, inherited. Pad size and substitute are the SKU-level variance. Steps 1 to 4 are analyst work, not an engineering project. Step 5 is where it scales to 40,000 SKUs or stalls at 200 hero products, because a one-off pass decays as the assortment turns over. ## Why is an empty conversational attribute a silent loss rather than an error? Because nothing turns red. An unpopulated optional attribute triggers no disapproval, no feed error and no drop in item count, so Monday's dashboard looks as healthy as before. The loss happens one layer down, in retrieval. A keyword surface degrades gracefully: a thin record ranks lower but still ranks. A question-scoped surface does not. When the shopper asks the underfloor-heating question, records that cannot answer are not ranked lower, they are not candidates. There is no report of a conversation you were not in, which is why feed health reporting cannot manage this surface at all. ## What are the failure modes? Four, in the order they happen. Each produces a populated field worse than an empty one. : Marketing language in a factual field. "Effortlessly versatile" occupies the slot and answers nothing. : Generating values from your own description. If the source never said whether the dress is lined, no model can produce that fact; it produces a plausible sentence, which is a misrepresentation risk. : Copy-paste across a category. "Requires a rug pad" is true for the 8x10 and wrong for the 2x3 runner. : Populating once. The assortment turns over continuously, so a set of answers written once describes a catalog you have stopped selling. ## What is still not documented, and what we will not guess Google announced these fields in May 2026, and several implementation specifics are not things we can state with confidence. We would rather say so than fill the gap. : Exact field names, cardinality and character limits are still maturing. Build against the current product data specification in Google's Merchant Center documentation, not a blog post, including this one. : Availability by country and category is not uniform, and we have seen no complete published matrix. : Attribute-level reporting. We are not aware of any. If Google ships it, this page gets a dated changelog entry. : How the Business Agent weights these fields against crawled page content when they disagree. Unknown; keep them consistent and watch Google's Shopping announcements. ## How do you know it worked when the surface has no reporting? You run a SKU-level holdout, because the surface will not tell you. Split a category into two cohorts matched on revenue, margin and traffic, populate the conversational layer on cohort A only, leave cohort B untouched for 28 days, and read difference-in-differences on revenue and product-page entrances, not impressions. That design cuts both ways, including against us. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control. That covers the product-language record as a whole, not this layer, so we will not call it a conversational-attributes benchmark until the cohort test finishes. Agentic checkout is still a rounding error next to the rest of ecommerce, so this is a cheap position on a surface still being built, while the same values sharpen the Shopping feed that pays this quarter. ## Where to start Pull last quarter's free-text return reasons for your top category, group them into the handful of questions that repeat, and count how many your records can answer. That baseline takes an afternoon. Then judge any approach, internal or bought, on whether it fixes the product-language input, runs continuously as catalogs churn, covers the surfaces that pay this quarter, and proves lift against a control. Would it survive a finance review? Start from Google's Merchant Center documentation. ## FAQ **What are Google Merchant Center conversational attributes?** Conversational attributes are Merchant Center fields Google introduced in May 2026 that let a product record answer what a shopper would ask a salesperson: what it works with, what to buy instead, and a product question's answer. They give AI Mode and the Business Agent something quotable, because the standard attribute set was built for keyword matching. **When did Google add conversational attributes to the Merchant Center?** Google announced the six Merchant Center conversational attributes at Google Marketing Live on 20 May 2026, four months after the Universal Commerce Protocol, the Business Agent in Search and Direct Offers in AI Mode arrived in January. They are one system: a protocol, a conversational surface, and the fields it reads. **Are conversational attributes required, or optional?** They are additional fields, not a replacement for the required product data specification, so leaving one empty produces no feed error and no disapproval. The cost is invisible rather than absent: the record is not a candidate when a shopper asks the question that field would have answered. **How is a conversational attribute different from a product description?** A description is one block of prose read as a whole, which forces an assistant to infer an answer and risk getting it wrong. A conversational attribute is a discrete, question-scoped value that can be retrieved and quoted directly, which is why literal values outperform marketing language here. **Where do I get the content for conversational attributes?** The answers are already written by customers in support tickets, free-text return reasons, product-page question threads, and reviews containing "wish I'd known." Mine those four at category level for the handful of questions that repeat, then vary the values at SKU level. **Do I need conversational attributes if I do not sell through an agentic checkout?** Yes, because these fields describe a product for conversational discovery rather than transaction, and most shoppers who research in an AI tool still buy on the retailer's own site. The same values sharpen onsite search and the Shopping feed, so the work pays on surfaces that already earn. --- # Merchant Center suspended for misrepresentation: the checklist, in the order we find them https://www.lily.ai/blog/merchant-center-misrepresentation-suspension 2026-08-25 Merchant Center suspended for misrepresentation? Work this checklist in the order we find them, catch every mismatch, and appeal once with the whole fix. A Merchant Center misrepresentation suspension almost never means Google thinks you are a fraud; it means something in your product record contradicts something else Google can see: price mismatches between feed and landing page, missing or contradictory policy pages, brand or GTIN conflicts, or claims in titles the page does not support. Google will not itemize the cause. Work the checklist in the order these causes most often turn up, fix everything found rather than the single most likely item, and appeal once with the full remediation described. If you are reading this with Shopping at zero, go to step 3. Everything below is ungated. ## What does a misrepresentation flag actually mean? Misrepresentation is Google's label for a gap between what your product record claims and what a shopper would find. It is a policy category applied to a data condition, not a judgement about your intent. Google builds its picture of an item from the feed, the landing page it crawls, its structured data, and the prices read through automatic item updates. When those disagree in a way that would change a shopper's decision, that is the violation under Google's Shopping ads policies. Item-level disapprovals name a reason and a SKU; account-level actions do not, because one label is covering many different conditions. ## Why won't Google tell you which product or which field? Google does not itemize misrepresentation causes, and support will typically restate the policy rather than name a SKU or a field. Plan an audit that does not need a diagnosis, because you are not getting one. So audit every mismatch class rather than the one you suspect. An account can be failing two at once, and a partial fix that leaves the second standing reads as a remediation that did not work. ## What triggers a Merchant Center suspension for misrepresentation? Four conditions account for nearly every case we audit, and this is the order we most often find them: price and availability disagreement between feed and landing page, thin or unreachable business and policy information, identity conflicts, and unsupported title claims. The notice reads identically in every case, so your signal is what changed just before it. One sentence covers six problems. Find the row matching when yours landed, and check that column first. What the notice says | What it points to | What to check first "Misrepresentation," no detail, account-level | Feed and page disagree on price or availability | price, sale_price, availability against the live page, best sellers first Same notice, mid-promotion | A sale ended on site but not in the feed | sale_price_effective_date, currency, tax and fee handling in both places Same notice, new store or domain | Thin business identity | Contact, returns and shipping pages: present, linked in navigation, crawlable Same notice, new brands or dropship | The record names one product, the page sells another | brand, gtin, mpn, identifier_exists, condition, whether link hits that variant Same notice, templated titles | Claims the page does not support | Shipping, price-superiority and authenticity language, against what the page states Same notice, luxury or resale | A brand relationship the site cannot evidence | Where the brand name appears, what the page says about sourcing, condition accuracy ## The checklist, in the order to work it Work these nine steps in sequence. Steps 3 to 5 cover the classes we most often find at fault; with one day, spend it there. 1. Freeze and export the feed first. Snapshot the exact file that was live when the suspension landed; you cannot rebuild it after a re-push. 2. Read the notice for scope and date. Confirm whether the action is account-level or item-level, then export every item-level issue from Products → Needs attention (Diagnostics in the older interface). Those are itemized, and they often point at the same condition. 3. Reconcile price and availability, feed against live page. For top sellers by clicks plus everything edited in the last 30 days, compare price, sale_price and availability to what the page renders now. Log every disagreement; fix nothing yet. 4. Check the structured data, not just the visible price. Google reads the Offer markup on the page, as set out in Google's product structured data documentation, so a correct on-screen price above a stale price, priceCurrency or availability is still a contradiction. 5. Walk the path to checkout and watch every number that moves. Mandatory fees, handling charges, currency switching, a tax treatment that differs from the feed. 6. Make the policy surface real. Shipping, returns, contact and business identity: findable from the homepage, crawlable, specific, consistent with your account settings. A returns page saying 30 days while the settings say 14 is a detectable contradiction. 7. Resolve identity conflicts. Check brand, gtin, mpn, identifier_exists and condition against the product actually on the page, and confirm link lands on that variant. Borrowed supplier GTINs turn up repeatedly in our audits; the product attributes working spec has the field detail. 8. Read your titles and descriptions as claims. "Free shipping," "authentic," "lowest price," condition words, warranty language. Either the page states it plainly or it comes out of the copy. 9. Fix everything, verify, then appeal once. Push the corrected feed, confirm a successful fetch, re-check a fresh sample against live pages, then appeal. The same discipline underpins product feed optimization. ## What if you sell luxury, designer or resale inventory? Luxury, designer and resale catalogs draw misrepresentation flags for a reason unrelated to price: the record implies a brand relationship the site cannot evidence. Make the relationship provable on the page rather than stripping the brand name and hoping. If you are an authorised retailer, say so and name the categories it covers. If you are a reseller or marketplace, describe the authentication process in detail and set conditions honestly instead of defaulting everything to new. Grey-market and parallel-import stock is hardest: the claim being read is not about quality but about your right to represent the item. Nobody writes this section because it is uncomfortable, and it resolves what everything else missed. ## How do you write the Merchant Center suspension appeal? Write one appeal, after remediation is complete and verified, structured as a record: what you found, what you changed, how you verified it, what prevents recurrence. Spend your effort on making the remediation complete rather than on polishing the wording; a reviewer is checking whether the account now complies, not how the sentences read. What we found. Every mismatched class is audited, including the clean ones: price parity, availability parity, Offer markup, checkout integrity, policy surfacing, identity fields, title claims. What we changed. Field level, with counts and dates. "Updated price and sale_price on \[n\] SKUs, corrected availability on \[n\], republished the returns page." How we verified. The sample re-checked, the method, the date of the last successful fetch, and confirmation that markup was checked too. What prevents recurrence. The monitor you put in place and who owns it. This is the block we most often find missing; it is what separates a real fix from a rushed one. Do not argue, do not ask which product was at fault, and do not submit mid re-crawl. ## How long does reinstatement take? Reinstatement runs through a review queue rather than an instant check, so it takes as long as the queue takes, and resubmitting does not shorten it. Repeat appeals buy nothing and cost you a clean record. While you wait, keep pushing the corrected feed so the fixed data is what gets re-crawled, move spend to surfaces still live, and hand finance the reconciliation report. ## How do you stop it happening again as the catalog churns? Suspensions recur because the conditions that caused them regenerate every time the catalog changes, and catalogs change daily. Treat agreement between record and page as a monitored invariant, not a project you closed in September. Watch four things continuously: price and availability parity, policy-page reachability, identity-field conflicts on new SKUs, and claim linting when titles are generated. A quarterly audit will not catch a promotion that expired on Tuesday. Shopping, paid social, onsite search and AI assistants read the same product-language record, which is why one contradiction takes down a revenue line and fixing the input once fixes it everywhere. That record has to be maintained continuously, because a catalog churning daily cannot be governed by a quarterly spreadsheet. The newer Merchant Center conversational attributes ride that same record, and the delivery layer is covered in product feed optimization. The check worth keeping is the one that survives a finance review. Pick ten SKUs at random tomorrow and show that record and page agree on price, availability, identity, and every claim in the title. If they do, and it happens again next Tuesday without anyone remembering, you are done. ## FAQ **What does "misrepresentation" mean in a Merchant Center suspension?** Misrepresentation is Google's label for a gap between what a product record claims and what a shopper would find, which in our audits is most often a feed-to-page price or availability disagreement. It describes a data condition, not a conclusion that the merchant is dishonest. **Can I get a Merchant Center suspension lifted without knowing the exact cause?** Yes, and that is the normal case, because account-level notices name neither the SKU nor the field. Audit every mismatch class in the order we most often find them, fix everything found rather than the likeliest item, then describe the remediation in one appeal. **How many times can I appeal a misrepresentation suspension?** Treat the first appeal as the one that counts and submit it only after remediation is complete and verified. Repeat appeals for partial fixes read as failed remediations, and resubmitting does not move you through the queue faster. **Does a price mismatch between my feed and my website really cause suspension?** Yes, and it is the trigger we find most often, including when the on-screen price is right but the Offer markup is stale. Google compares feed, rendered page and markup, so all three must agree through checkout. **Do my shipping and returns policies need to be on the site itself?** In practice yes: findable from the homepage, crawlable rather than JavaScript-only, specific about timeframes and costs, and consistent with your account settings. A returns page contradicting the configured return window is detectable without a human. **Why was my account suspended when all my products were approved?** Item-level approval and account-level compliance are separate checks, so a catalog of approved products can sit under a suspended account. Account-level misrepresentation points at business information, policy pages or a catalog-wide pattern rather than any single item. --- # Product attributes: which ones each discovery surface reads, and what a good one looks like https://www.lily.ai/blog/product-attributes-101 2026-08-25 Which product attributes each discovery surface reads, what a well-formed value looks like, and the 2026 Merchant Center fields to fill in first. Product attributes are the structured facts that describe a product (material, fit, occasion, color family, compatibility) and they are the unit of language every discovery surface reads. Google Shopping matches queries against them, Meta Advantage+ builds audiences from them, onsite search filters on them, and AI shopping assistants use them to decide whether your product answers a shopper's question. The attributes that drive discovery are usually not the ones in the PIM, because merchandisers catalog by SKU logic and shoppers search by use. This page is a working spec, not an introduction. It lists the six attribute types, the 32 fields that decide whether a product is retrievable at all, what each surface does with them, what a well-formed value looks like next to a badly formed one, and why the whole set decays on a schedule nobody at the retailer controls. It is written to be kept open in a tab while you audit a category. ## What is a product attribute, exactly? A product attribute is a single structured fact about a product, stored as a named field with a value: material: linen, size_system: US, occasion: wedding guest. One fact, one field, one value: that constraint is the whole discipline, because a field holding two facts is a field no machine can filter on. Three things get called attributes and are not. Taxonomy is where the product sits in a hierarchy, not what it is made of; it decides which queries you are eligible for, and it gets its own treatment in product taxonomy. Description is prose, and prose is read but not filtered. Specification is the manufacturer's version of an attribute, written for a spec sheet rather than a shopper, which is why so many catalogs contain fabric: 100% CO where the shopper typed "cotton." The practical test for whether something is a real attribute: could a shopper narrow a result set with it, and could a machine sort on it without guessing? If both answers are yes, it belongs in a field. If not, it belongs in the description. ## What are the six types of product attributes? There are six types, and they fail in six different ways: identity, physical, descriptive, contextual, style, and compatibility. Most catalogs are strong on the first three (those are the fields a PIM was built to hold) and close to empty on the last three, which is exactly where discovery is decided. ### Identity attributes Identity attributes establish that this record refers to a specific, real, buyable product: id, gtin, mpn, brand, item_group_id, condition, google_product_category, product_type, gender, age_group. When they are wrong the item is not ranked badly, it is disqualified: duplicate GTINs, missing item_group_id on a variant family, and a google_product_category that does not match the item are three of the most common quiet suppressors in Merchant Center. ### Physical attributes Physical attributes describe the object itself: size, size_system, size_type, color, material, pattern, plus dimensions and capacity. These are the ones onsite facets are built from, which means an unpopulated physical attribute does not just lose a query, it removes the product from a filter a shopper has already chosen. ### Descriptive attributes Descriptive attributes are the written surface of the record: title, short_title, description, product_highlight, product_detail. The product_highlight field in Merchant Center is one of the most under-used fields in retail: it takes short, benefit-shaped bullets and it is the cheapest place to put the facts that have no dedicated field of their own. ### Contextual and occasion attributes Contextual attributes state when, where and why a product is used: occasion, season, use case or activity, room or setting, weather suitability. Almost no catalog carries them as fields, and almost every natural-language shopping query contains one ("wedding guest," "for a small apartment," "for humid weather") which is why this is the highest-yield gap on this list. ### Style attributes Style attributes describe how a product reads aesthetically: style or aesthetic, silhouette or shape, fit and drape. Size and fit attributes deserve their own attention in apparel because they carry two jobs at once, retrieval and returns: "true to size, relaxed through the hip" is both a match signal and a returns-prevention signal. ### Compatibility attributes Compatibility attributes answer whether the product works with something else the shopper already owns: works-with, and substitutes. These were homeless until May 2026, when Google added conversational attributes to Merchant Center covering compatibility and alternatives. ## Which product attributes does each discovery surface read? Every surface reads the same record, but each one does a different job with it. That is the reason attribute work compounds across channels and formatting work does not: one improvement to the product-language input is read four times. This table maps each attribute type to what the four discovery surfaces actually do with it. Attribute type | Google Shopping / PMax | Meta Advantage+ | Onsite search & filters | AI shopping assistants Identity | Resolves the item, controls eligibility, handles variants | Keys the catalog to the product set and the pixel | Collapses variants under one PDP | Decides whether the record resolves to a real, buyable item Physical | Query-to-attribute matching on size, color, material | Source material for audience and segment building | Powers the facets shoppers click | Answers the literal constraint in the prompt ("linen," "size 14") Descriptive | Read from title, short_title, description, product_highlight; drives impression share | Supplies ad copy and creative variants | Feeds the free-text index behind the search box | The main source of quotable language Contextual / occasion | Matched on long-tail queries and, since 2026, conversational attributes | Seasonal and occasion segments | Rarely a facet, frequently a query | The biggest gap: most prompts are occasion-shaped Style | Matched on aesthetic and silhouette terms | Creative and audience targeting | Facets in apparel and home | Matched against descriptive, adjective-heavy prompts Compatibility | Covered by the conversational attributes introduced May 2026 | Cross-sell product sets | Accessory and "works with" modules | Determines whether the item survives "will this work with X" The uncomfortable part of this table is the last column. AI assistants do not browse a catalog and form an opinion; they retrieve against records, so a product whose record never states occasion or fabric weight is not ranked lower, it is not a candidate. ## Which 32 attributes actually decide whether you show up? These are the 32 fields we check first on any catalog audit, grouped by type, with a note on whether the field exists natively in the Merchant Center. Twenty-one of the 32 map to a long-standing Merchant Center attribute, two arrived as conversational attributes in May 2026, and the remaining nine have no field of their own, which is precisely why they get skipped, and precisely why they are available. This is the reference list: every field, its type, whether Merchant Center has a native home for it, and the surface it moves the hardest. # | Attribute | Type | Native field? | Moves hardest on 1 | id | Identity | Yes | All surfaces 2 | gtin | Identity | Yes | Shopping eligibility 3 | mpn | Identity | Yes | Shopping eligibility 4 | brand | Identity | Yes | Shopping, AI assistants 5 | item_group_id | Identity | Yes | Variant handling everywhere 6 | condition | Identity | Yes | Shopping, marketplaces 7 | google_product_category | Identity | Yes | Shopping eligibility 8 | product_type | Identity | Yes | Onsite, Shopping 9 | gender | Identity | Yes | Apparel Shopping 10 | age_group | Identity | Yes | Apparel Shopping 11 | size | Physical | Yes | Onsite facets, Shopping 12 | size_system | Physical | Yes | Cross-market Shopping 13 | size_type | Physical | Yes | Apparel Shopping 14 | color | Physical | Yes | Onsite facets, AI prompts 15 | material | Physical | Yes | AI prompts, Shopping 16 | pattern | Physical | Yes | Onsite facets 17 | Dimensions and capacity | Physical | No, use product_detail | Home, furniture, AI prompts 18 | title | Descriptive | Yes | Shopping impression share 19 | short_title | Descriptive | Yes | Shopping surfaces, AI Mode 20 | description | Descriptive | Yes | Onsite search, AI retrieval 21 | product_highlight | Descriptive | Yes | Shopping, AI retrieval 22 | product_detail | Descriptive | Yes | Onsite facets, AI retrieval 23 | Occasion | Contextual | No | AI assistants, long-tail Shopping 24 | Season | Contextual | No | Advantage+, seasonal queries 25 | Use case or activity | Contextual | No | AI assistants, onsite 26 | Room or setting | Contextual | No | Home category, AI assistants 27 | Weather or temperature suitability | Contextual | No | Apparel, AI assistants 28 | Style or aesthetic | Style | No | Onsite facets, AI prompts 29 | Silhouette or shape | Style | No | Apparel onsite, AI prompts 30 | Fit and drape | Style | No | Apparel conversion and returns 31 | Works-with / compatible accessories | Compatibility | Conversational, 2026 (name not published) | AI assistants, Shopping 32 | Substitutes | Compatibility | Conversational, 2026 (name not published) | AI assistants, Shopping The nine homeless attributes do not stay homeless. They go into product_detail as name/value pairs, into product_highlight as short bullets, and into the description in shopper phrasing, which means the retailer who does this deliberately is populating fields the rest of the category is leaving blank. Delivery mechanics for all of it sit in product feed optimization. ## What does a well-formed attribute value look like? A well-formed attribute value holds exactly one fact, in the vocabulary a shopper would use, in a unit a machine can parse, with no marketing language attached. A badly formed value usually fails on one of four things: it is internal shorthand, it bundles two facts, it is missing its unit or system, or it is a null dressed up as a string. Same product, same field, two versions: the left column is what audits usually find, the right is what the surfaces can actually use. Field | Badly formed | Well formed | What breaks color | MULTI / AST / Cypress | Sage Green with color family Green | Color facets return nothing; "sage green dress" never matches size | M | M, size_system: US, size_type: regular | Cross-market Shopping mis-serves; onsite size filter is unreliable material | 95% PL 5% EA | Linen, with product_detail fabric weight Midweight | "Linen dress" and "breathable fabric" queries never retrieve it title | SS26 LN MIDI DR SGE 8 | Linen Midi Dress, Sage Green, Women's - Sleeveless | Impression share on the highest-intent Shopping queries product_highlight | Great for any occasion! | Breathable linen for hot weather, Cut for outdoor weddings | Nothing is retrievable; "any occasion" matches no occasion Fit | Regular | Relaxed through the hip, true to size | Fit queries miss, and returns stay high Works-with | Fits most models | Fits iPhone 15 and 15 Pro. Not compatible with 15 Plus. | Assistant answers "will this fit" with a guess or a competitor Occasion | (empty) | Wedding guest, Warm-weather event, Daytime formal | The entire occasion query set, which is most AI prompts Two rules that catch most of it. First, if a value would need a glossary to interpret, it is a code, not an attribute. Second, if a value could be pasted onto any product in the catalog without becoming false, it carries no information: "great for any occasion" is the canonical example. ## Product attribute examples by category: apparel, beauty, home The gap between catalogd and searched looks different in every category, and the fix is category-specific. Here is what the missing layer usually is in three. ### Apparel Catalogd: style number, season code, fabric composition, colorway name, size run. Searched: garment type, fabric in plain words, occasion, weather, sleeve length, rise or hemline, fit. The attributes that move apparel are the size and fit attributes plus occasion, "wedding guest," "true to size," "midweight linen", and none of them arrive from the vendor. ### Beauty Catalogd: SKU, shade code, volume, ingredient list, product family. Searched: shade in plain description, undertone, coverage level, finish, skin type, skin concern, whether it is fragrance-free. Shade code to shade language is the single highest-yield mapping in beauty, because N-24 retrieves nothing and light neutral with a golden undertone retrieves the query the shopper actually typed. ### Home Catalogd: item number, dimensions, material, collection name, vendor. Searched: room, style, size in relatable terms, washability, pet and child suitability, assembly requirement. Home is where dimensions need translation: 243 x 152 cm is a fact, and fits under a queen bed with 18 inches of walking space is a fact a shopper can act on, so the record needs both. ## What are Merchant Center conversational attributes, and what goes in them? Conversational attributes are Merchant Center fields Google introduced at Marketing Live on 20 May 2026, four months after the Universal Commerce Protocol, the Business Agent and Direct Offers in AI Mode, that let a product record answer the questions a shopper would ask a salesperson rather than only state specifications. They exist because the standard attribute set was written for keyword matching, and the surfaces reading it now hold conversations. What actually goes in them is shopper-question language, not catalog language. Compatibility answers take specific model names and explicit exclusions. Alternatives take the honest answer to "what should I get instead if this is out of stock or out of budget." The rest take the answers your support team already types every week, which is where the content comes from, and it is free. Google has not published the full field names yet, so carry that language in product_detail and product_highlight until it does. The field-by-field version is in Merchant Center conversational attributes. ## Where do the attributes in your PIM and the attributes shoppers use diverge? They diverge almost completely, because the two systems were built by different people for different jobs. A PIM record is organized so a business can buy, price, allocate and reconcile a product; a query is organized around what a person wants to do with it. One worked example, not a survey: a single linen midi dress, its populated PIM fields against the attributes its queries depend on. What the PIM carries (14 populated fields) | What the shopper searches on (8 attributes) Style number, season code, buying department, supplier code, country of origin, fabric composition, care code, colorway name ("Cypress"), size run, cost, retail price, launch date, image asset ID, GTIN | Garment type ("midi dress"), fabric ("linen"), color in plain words ("sage green"), occasion ("wedding guest"), season ("summer"), sleeve ("sleeveless"), fit ("relaxed"), price band ("under $300") Three attributes appear on both lists: garment type, fabric and color. All three fail on the value rather than the field: garment type is implied by a taxonomy node rather than stated, fabric is a composition code, and the color is "Cypress." The record is complete and unusable at the same time, which is the failure mode covered in product data enrichment. Nothing in that list is the merchandiser's mistake. The fields were filled in exactly as the system asked; the system was designed to move stock through a warehouse, not to answer a question typed into a search bar at 11pm. ## Why do product attributes decay, and what does product attribute management involve? Attributes decay because a catalog is a population, not a project: SKUs churn, vendors change, and every drop arrives with vendor-supplied data written in vendor vocabulary. A one-time enrichment pass is accurate on the day it ships and less accurate with every drop, vendor swap and rename that lands after it, which is why attribute quality decays rather than completes. Four decay sources, in the order they usually bite: 1. New SKU intake. Every new item lands with supplier data. Unless enrichment runs on intake, the newest products (the ones with the most demand) have the worst records. 2. Vendor and colorway changes. The same product returns next season with a different composition string and a renamed colorway, and every mapping built against the old value quietly stops matching. 3. Platform change. Field requirements move. Google's 2026 additions created fields that did not exist when most catalogs were last enriched, and a field that does not exist cannot be populated retroactively by a project that ended in 2024. 4. Shopper vocabulary drift. The words people use for style, fit and occasion move faster than any catalog process. Values that matched last year's language stop matching this year's. Product attribute management, done properly, is therefore a running process with a trigger on catalog change, not a quarterly initiative. That is criterion (b): it has to run continuously, because catalogs churn. ## How do you know attribute work moved revenue? You know because you held out a control group, matched the spend, and read the difference: anything else is attribution, not evidence. Platform-reported lift is not a control; the platform is grading its own homework. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control group. On Meta Advantage+, the same input change returned +21.4% ROAS against a holdout, cross-validated with Meta's own Conversion Lift. Onsite, a statistically significant A/B test returned +28.3% onsite revenue. We ran 1,000-plus controlled tests before publishing the benchmark those figures come from. The reason attribute work is measurable at all is that it is upstream of every surface: the same record change can be read on Shopping, on Advantage+, on the search box and in an assistant's answer. Set the test up before the rewrite, not after, and keep a slice of the catalog untouched for the full window even though it will feel wasteful. The surface-level version of this measurement is in product feed optimization. One market number worth holding next to all of this: Semrush found that 22% of US shoppers have bought inside an AI tool while 50% bought elsewhere after researching one. The assistant is mostly a discovery surface, not a checkout, and the thing it discovers you by is the attribute record. ## What to ask before you buy attribute work Ask four questions, and ask them in this order. They are the same four that decide whether the work survives contact with a finance team. 1. Does it fix the product-language input, or the layer above it? A dashboard that reports attribute coverage is not attribute work. 2. Does it run continuously? A one-time pass is accurate the day it ships and decaying from the next catalog change onward. 3. Does it cover the surfaces that pay this quarter? Paid surfaces read the same record. If the work only touches onsite, most of the value is left on the table. 4. Is the lift proved against a control? Ask for the holdout design, the window, and what the result does not prove. Compressed to one sentence: would it survive a finance review? Start by pulling one category, running the 32-field table against ten SKUs, and counting how many of the nine homeless attributes are populated anywhere in the record. The number you get is your ceiling on every surface at once. Delivery sits in product feed optimization, placement in product taxonomy, and the upstream job in product data enrichment. Primary sources: *Google's Merchant Center product data specification* and the *Google product taxonomy*. ## FAQ **What are product attributes in ecommerce?** Product attributes are the structured facts stored about a product in named fields, such as material, color, size, occasion and compatibility. They are what discovery surfaces read to decide whether a product matches a query, so a fact that is not in an attribute field is a fact no machine can filter, match or quote. **What are the different types of product attributes?** There are six practical types: identity, physical, descriptive, contextual or occasion, style, and compatibility. Most catalogs are well populated on identity, physical and descriptive attributes because those are the fields a PIM was built to hold, and thin on contextual, style and compatibility attributes, which is where natural-language demand now sits. **What is the difference between a product attribute and a product taxonomy?** An attribute is a fact about the product, and a taxonomy is the position of the product in a hierarchy. Taxonomy decides which queries a product is eligible to appear for at all, and attributes decide whether it wins among the products that are eligible. **Which product attributes matter most for Google Shopping?** Identity attributes decide eligibility, so gtin, brand, item_group_id, google_product_category and condition have to be correct before anything else matters. After that, the title, short_title, product_highlight and the physical attributes (color, size with its size system, material and pattern) do the most work on impression share for high-intent queries. **How many product attributes should a product have?** There is no target count, because a catalog can be 100% populated and still fail to retrieve if the values are internal codes. A more useful test is whether the record can answer the six or eight things a shopper would actually specify for that category, including occasion, fit and compatibility, in the words a shopper would use. **How often should product attributes be updated?** Attribute quality decays continuously because catalogs churn, so the useful cadence is triggered by catalog change rather than by calendar. New SKU intake, vendor or colorway changes, and platform field changes each need enrichment to run at the moment they happen, not at the next quarterly project. --- # Product Data Enrichment: What It Means When the Machines Are the Reader https://www.lily.ai/blog/product-data-enrichment 2026-08-25 Product data enrichment is not filling empty fields. Get the working definition, the four readers, the boundary with PIM, and a way to measure lift. Product data enrichment is the process of adding and improving the attributes, descriptions and taxonomy attached to each product so that the record is complete and legible to the systems that read it. Completeness alone does not produce discovery: a catalog can have every field populated and still fail, because the fields are written in merchandiser vocabulary rather than shopper vocabulary. Enrichment that moves revenue changes what the record says, not just how much of it is filled in. That is where the working definition and the market's part company. Most enrichment programs are scoped and reported against fill rate, which measures how much of the record exists, not whether anything reading it can understand it. This page defines the term, separates it from the jobs it gets confused with, and gives you a way to tell whether a program moved revenue or moved a chart. ## What does product data enrichment actually cover? Product data enrichment covers three assets: the structured attributes on a SKU, the copy that describes it, and the taxonomy that places it. Everything else sold under the word is a prerequisite (cleansing, normalization) or a downstream step (syndication). (Attributes) material, fit, occasion, heel height, compatibility, care. These are the units every surface matches a query against, covered field by field in our working spec for product attributes. (Descriptions) enriched copy states the facts a shopper rules the product in or out on, in the first two sentences, in their words. (Taxonomy) where the product sits, and which external hierarchies it maps to. A misplaced product is invisible even when perfectly described, which is why product taxonomy belongs inside enrichment. Enrichment has two halves, routinely treated as one. Coverage enrichment fills an empty field: material was blank, now it reads full-grain leather. Language enrichment rewrites a populated field in the reader's vocabulary: product_type reads FW26 / Boots / BT8812, now it reads Women's Chelsea Boots. Everyone ships the first half. Revenue moves on the second. Ship coverage alone and you get an excellent completeness chart against a flat revenue line. ## Why does a 100%-complete catalog still fail to show up? Because a completeness score grades the record against a schema, and the schema was written by the people who own storage, not the people who search. A record can satisfy every required field in Google's Merchant Center product data specification and contain none of the words a shopper types. The same boot, twice. The left column passes a completeness audit at 100%; the right is the same product described the way it is searched for. Field | At 100% completeness | After language enrichment title | Ladies Boot BT8812 Blk | Women's Black Leather Chelsea Boot, Waterproof, Wide Fit product_type | FW26 > Footwear > BT8812 | Women's Shoes > Boots > Chelsea Boots material | LTHR-FG | Full-grain leather description | Style BT8812 from the FW26 collection. | Waterproof full-grain leather Chelsea boot, elastic side panels, 3.5 cm block heel. Wide fit. occasion | (not in the schema) | Everyday, work, rain Nobody did anything wrong on the left. It is what a diligent team produces when handed a schema, a deadline and a fill-rate target. Completeness was the right goal when a human read the page; when the reader is a ranking system, the goal is legibility. ## Enrichment, cleansing, normalization, syndication: what is the difference? They are four separate jobs and only one of them changes what the record says. Cleansing removes errors from a record that already exists, normalization converts values to one consistent format and unit, enrichment adds facts the record never carried and rewrites the ones it carries badly, and syndication delivers the finished record to each destination. Four jobs frequently sold under one word, and the hard limit of each. Job | What it does | What it cannot do Cleansing | Removes errors, duplicates and contradictions from existing records | Add anything never captured Normalization | Converts values to one format, unit and controlled vocabulary | Make a consistent term the right term Classification | Maps each product to a node in an internal and external taxonomy | Improve how the product reads once found Enrichment | Adds missing attributes and rewrites existing ones in the reader's vocabulary | Guarantee the record reaches the surface Syndication | Transports the finished record to each channel in that channel's schema | Improve a record that was weak before it left The adjacent vocabulary matters, because buyers pay for the overlap twice. Product information enrichment and catalog enrichment are the same work at different scopes. Entity resolution, matching and deduplication are cleansing operations that produce a golden record. Attribute extraction proposes values from copy and imagery; it is an input to enrichment, not enrichment. Validation is the pre-flight check against a destination schema. The working order is cleanse → normalize → classify → enrich → validate → syndicate. Enrichment sits fourth, it is the only step that adds meaning rather than order, and it is the only one no platform performs for you. ## Which readers are you enriching for, and what does each one need? Four systems read the product record before a shopper sees it, and each reads a different part of it first. Enriching for one and assuming the rest follow is how good programs underperform. What each discovery surface reads first, and what fails when the record is thin. Reader | Reads first | What "enriched" means to it | Failure mode Google Shopping | title, product_type, google_product_category, attributes | Query vocabulary in the title, in the documented attribute order, plus the conversational attributes added in 2026 | No impressions, and no error to debug Meta Advantage+ | attributes and description, for audience construction | Attributes rich enough to find look-alike demand | Broad targeting on a vague record; ROAS caps out Onsite search | attributes as facets, description as match text | Facets shoppers actually filter on, in their words | Null results on products you stock AI assistants | the whole record as prose, judged on whether it answers a question | Facts stated explicitly, not implied by a category or an image | The assistant recommends whichever product said so plainly That last row changed most recently. Google announced the Universal Commerce Protocol and a Business Agent inside Search in January 2026, and Amazon folded Rufus into Alexa for Shopping in May 2026. Semrush found that 22% of US shoppers have bought inside an AI tool and 50% bought elsewhere after researching in one: the assistant is a recommendation surface long before it is a checkout surface, and a recommendation is made out of language. All four read the same record. The platforms built every pipe, and not one of them supplies the quality of the input. ## Where does enrichment end and the PIM or feed manager begin? A PIM is the system of record and a feed manager is the transport layer; neither writes better language, and neither claims to. The boundary matters because buyers routinely purchase storage or transport and book it as enrichment. A PIM holds the fields, governs who may edit them and keeps versions straight. It is a filing cabinet with excellent labels, and it reports 100% fill rate because that is a true fact about the cabinet. A feed manager maps what is inside to each destination schema, applies rules and ships it. Neither decides that LTHR-FG should read full-grain leather for Google Shopping and waterproof leather when the shopper's question is about rain. Enrichment is not a PIM, not a feed manager and not a DAM; it is the content those systems store and move. If the demo is a field-mapping screen you are being shown transport; if it is a completeness dashboard you are being shown storage. The second boundary is the expensive one. Enrichment delivered as a one-time agency project (a rewrite sprint, a deliverable, a handover) fails the continuity criterion the moment the next assortment lands. The work is not bad; the share of the catalog carrying that language falls with every new SKU that lands, and the invoice does not. ## Why does catalog enrichment have to run continuously? Because a catalog is a population, not a project, and it churns underneath any finished deliverable. New SKUs arrive un-enriched, assortments turn over, categories get restructured, and destination schemas move: Google added six Merchant Center conversational attributes in May 2026, and every catalog enriched before that date became incomplete against the new specification overnight. Shopper vocabulary drifts as well, faster than catalogs do, and no static rewrite tracks it. So enriched coverage decays from the day it ships, invisibly, because the new SKUs are complete. Complete and illegible. Continuous enrichment holds the newest SKU to the same language standard as the flagship one, without anyone scheduling a project to make it so. ## How do you measure product data enrichment? Against a control, or you have not measured it. Product data quality gets no honest read from a completeness score, a disapproval count or a platform's own attribution: the first two describe the record, and the third grades its own homework. What each measurement method can and cannot support in front of a finance team. What you measured | What it proves | What it does not prove Fill rate / completeness score | The fields contain values | That anything can read them Disapprovals cleared | The record is admissible | That it is competitive Impression share | The record now appears | That the appearance sold anything Platform-reported lift | The platform is confident | Incrementality; it attributes to itself Holdout with matched spend | Revenue difference against an untouched control | Anything outside the test population The design that survives review is unglamorous. Split the catalog into a treatment group and an untouched holdout, match spend across both, hold bids, budgets, creative and promotions steady, and read the difference in differences over a full purchase cycle. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control. On Meta Advantage+, a holdout cross-validated with Meta's own Conversion Lift returned a 21.4% ROAS improvement. Onsite, a statistically significant A/B test returned a 28.3% increase in onsite revenue. Those figures sit on top of more than 1,000 controlled tests run before the benchmark was published. The full design, and what the results do not prove, is in does feed optimization increase sales. eMarketer put the share of US brand and agency marketers using incrementality testing at 52% in July 2025, so about half the market now reads enrichment against a control and the other half is still buying it on a fill-rate chart and calling that evidence. Enrichment that is not measured against a control fails the proof criterion, however good the language is. ## What does an enrichment program that survives a finance review look like? It changes the input rather than the layer above it, runs continuously, covers the surfaces that pay this quarter, and proves lift against a control. Those four compress into the question a CFO will ask anyway: would this survive a finance review? Use them on any vendor, including us: 1. Does it change the product-language record itself, or re-map what you already have? 2. Does it run on every new SKU automatically, or hand over a deliverable that starts decaying? 3. Does it cover Google Shopping, Meta Advantage+, onsite search and AI assistants, or only the surface that demos best? 4. Will the vendor design a holdout with you, name the control group, and accept the difference-in-differences read? If the answer to (2) is a statement of work with an end date, or the answer to (4) is a platform dashboard, you are buying a project rather than an input fix. Do this instead, take ten SKUs from your best-selling category and mark every field a shopper would search on that is missing or written in your own vocabulary. That costs an afternoon, it tells you what the completeness score cannot, and it is the first evidence you will hold that would survive a finance review. ## FAQ **What is product data enrichment?** Product data enrichment is the process of adding and improving the attributes, descriptions and taxonomy attached to each product so the record is complete and legible to the systems that read it. It differs from filling empty fields because the target is language a ranking system can parse, not a higher completeness score. **What is the difference between product data enrichment and data cleansing?** Data cleansing removes errors, duplicates and contradictions from information a record already contains. Enrichment adds facts the record never carried and rewrites weak values in the reader's vocabulary, so cleansing improves accuracy while enrichment changes what the record says. **Does a PIM do product data enrichment?** A PIM stores, governs and versions product attributes, which makes it the system of record rather than the system that improves the language inside it. It will report a catalog 100% complete while every title is still written in internal merchandising vocabulary. **How do you measure the ROI of product data enrichment?** Split the catalog into a treatment group and an untouched holdout, match spend across both, and read the difference in differences. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the untouched control. **How often should product data be enriched?** Continuously, because catalogs churn: new SKUs arrive un-enriched, assortments turn over, and destination schemas change, as Google's Merchant Center conversational attributes did in May 2026. A one-time project starts decaying on handover, and fill-rate reporting hides it because the new SKUs are complete. **What should I look for in product data enrichment tools?** Judge any tool on four things: does it change the product-language record itself, does it run on every new SKU, does it cover Google Shopping, Meta Advantage+, onsite search and AI assistants, and will the vendor design a holdout with you. If the evidence offered is a platform-reported lift number rather than a control group, the claim cannot be checked. --- # Product feed optimization: what it is, what actually moves revenue, and how to prove it https://www.lily.ai/blog/product-feed-optimization 2026-08-25 Product feed optimization is a vocabulary problem. The three layers, the eight changes that move revenue, and how to prove the lift with a control. [image] Product feed optimization is the work of improving the product-language record (titles, attributes, descriptions and taxonomy) that Google Shopping, Meta Advantage+, onsite search and AI assistants all read before deciding whether to show your product. Most feed work fixes formatting errors, but the lift comes from the vocabulary layer, because the feed is scored against how shoppers describe products, not how the merchandiser catalogd them. In a matched-spend A/B test with a 28-day holdout, rewriting the product-language input produced a 28% increase in Google Shopping revenue against the control. That is the argument. The rest of this page is the mechanism. (Spelled product feed optimisation outside the US. Same job, same fields.) ## What is a product feed, and who actually reads it? A product feed is a structured file of product records, one row per item, one column per attribute, handed to a platform so it can decide when your product is a good answer to somebody's intent. A feed is the only sentence your product gets to say, and four machines read that sentence for four different reasons. The feed is not a Google artifact. The same record feeds every discovery surface you sell on. Every discovery surface reads the same product-language record. Here is what each one takes from it. Surface | What it reads first | What it does with it Google Shopping / PMax | title, google_product_category, product_type, filterable attributes | Matches the query to the item and sets the auction's relevance floor Meta Advantage+ | title, description, product_type, variant structure | Builds the interest signal that targets the catalog Onsite search | Attributes, taxonomy, descriptors | Powers facets, autocomplete, zero-result recovery AI assistants and AI Mode | The full record plus the product page | Decides whether the product can satisfy a constrained request The platforms built the pipes between these surfaces in under a year: the Agentic Commerce Protocol in September 2025, Google's Universal Commerce Protocol on 11 January 2026, then Universal Cart and the six Merchant Center conversational attributes in May 2026. Not one of them supplies the quality of what flows through the pipe. That part is still yours. ## What are the three layers of product feed optimization? Feed work stacks in three layers, and they are not equally valuable. Syntax fixes disapprovals, structure fixes routing, and language fixes whether you were eligible for the query at all. Most product feed best practices stop at layer one and call it a strategy. The three layers of feed work, what each fixes, and what each is worth. Layer | The work | What it fixes | Ceiling 1 · Syntax | Errors and disapprovals, valid GTINs, price and availability parity | Whether the item is allowed to serve | Gets you to zero, never above it 2 · Structure | google_product_category, product_type, variant grouping, custom labels | Where the item is routed and how it can be bid on | Real, usually a one-time gain 3 · Language | Titles, descriptors, attribute values, the vocabulary shoppers use | Whether the item matches the intent at all | Where the compounding revenue is Layer one is hygiene. A clean feed with the wrong vocabulary is a well-formatted product nobody can find. That is the trap. Every diagnostic tool reports on layer one, because that is the layer that produces error codes. Nothing in the Merchant Center throws a warning that says your title uses a word your buyer never types. None of this is the merchandiser's fault. They catalogd the product correctly for the system they were handed, a PIM built to track inventory, not to answer questions. The field pointed a discipline at the layer that generates alerts, and the revenue chart stayed flat. ## Where does product feed work fail most often? The same failures show up in nearly every catalog audit, and they cluster in the language and structure layers rather than syntax. Ordered by how often we find them, most frequent first. The feed failures we find most often, ordered by frequency, with the surface each one hurts most. # | Failure mode | What it costs | Hurts most 1 | Title in internal style: SKU codes, house color names, brand first | Never enters the set for descriptive queries | Shopping, onsite 2 | Filterable attributes empty (material, pattern, size_system, age_group) | Filtered out of every refined search | Shopping, onsite 3 | google_product_category auto-assigned to a top-level bucket | Benchmarked against the wrong products, so the bid is wrong too | Shopping, PMax 4 | product_type left as a two-level internal path | No taxonomy for bidding to act on | PMax, Advantage+ 5 | Description opens with care instructions and legal copy | The useful sentence sits below where machines read | Shopping, AI assistants 6 | product_highlight and product_detail entirely empty | The richest fields in the spec ship blank | AI assistants 7 | Variants ungrouped: no item_group_id | The product competes against itself | Shopping, Advantage+ 8 | custom_label_0 to 4 undifferentiated | Bidding cannot act on margin, velocity or season | PMax, Advantage+ 9 | Feed refreshed on schedule, language never re-scored | New SKUs undo the last pass | All four Number nine is the one that turns a successful project into a flat year, and it is why continuity belongs in the buying criteria rather than the feature list. ## What does a rewritten product record actually look like? Here is one anonymized record from a mid-market apparel catalog, before and after the language layer was rewritten, field by field. Nothing about the product changed. Only the words it uses. One product record, before and after the language rewrite. Same product, same price, same image. Field | Before (as it left the PIM) | After (rewritten against shopper vocabulary) title | AVERY DRESS 2208-BLK-M | Avery Unlined Linen Midi Dress, Black, Square Neck, Sleeveless description (first line) | Avery Dress. Style 2208. Machine wash cold. Imported. | An unlined black linen midi dress with a square neck and no sleeves, cut straight through the waist and true to size. google_product_category | Apparel & Accessories | Apparel & Accessories > Clothing > Dresses product_type | Women > Dresses | Women > Dresses > Midi Dresses > Linen > Sleeveless color | BLK | Black material | (empty) | Linen pattern | (empty) | Solid size / size_system / size_type | M / (empty) / (empty) | M / US / regular age_group / gender | (empty) / (empty) | adult / female product_highlight | (empty) | Unlined, so it wears cool in heat · Falls mid-calf on 5'6" · Runs true to size product_detail | (empty) | Fit: relaxed through waist · Lining: unlined · Neckline: square item_group_id | (empty) | AVERY-LINEN-MIDI custom_label_0 | SS26 | margin_high Read the "before" column the way a machine reads it. Not one word in it is a word a shopper types. "Unlined" is missing, and no standard Merchant Center attribute holds it, which is why it has to travel in the title, description, product_highlight and product_detail, fields that Google's product data specification defines and most catalogs ship empty. The title attribute accepts 150 characters. This one used 22 of them, ten of which are a SKU fragment. The rewrite invented nothing. Every added word is a checkable property that already existed somewhere (the spec sheet, the sample notes, the returns data) and had simply never reached the record. That is the product data enrichment job, and the feed is where it gets delivered. ## How do you optimize a product feed? The eight changes, in order Run them in this order. Steps 1 and 2 gate everything downstream, steps 3 to 7 are the language layer where the revenue is, and step 8 lets media buying act on the work. The eight changes, the order to run them in, and the surface each one moves. Step | The change | Surface it moves 1 | Clear all disapprovals and price/availability mismatches | All four (eligibility gate) 2 | Set google_product_category correctly and build a deep product_type path | Shopping, PMax 3 | Fill every filterable attribute: color, size, size_system, material, pattern, age_group, gender | Shopping, onsite 4 | Rebuild titles against shopper vocabulary, most-searched descriptor first | Shopping, PMax, onsite 5 | Rewrite the opening of the description to answer the shopper's question | Shopping, AI assistants 6 | Populate product_highlight and product_detail | Shopping, AI assistants 7 | Carry the vocabulary no standard field holds: fit, occasion, feel, use case | AI assistants, onsite 8 | Group variants with item_group_id; segment custom_label_0 to 4 by catalog economics | PMax, Advantage+ A few of these get misread. Step 2 is a taxonomy decision, not a dropdown: google_product_category sets the competitive set you are benchmarked inside, and a top-level assignment drops a linen midi dress into the same pool as socks (product taxonomy has the argument). Step 4 is where good intentions do the damage, because a title optimized for brand consistency and a title optimized for retrieval are different objects, and only one of them is scored by a machine. Step 7 has no error code attached to it, and it is exactly what the six May 2026 conversational attributes were built to receive. The field-level reference for steps 3 to 7 is product attributes. ## How is Google Shopping feed optimization different? Google Shopping feed optimization is the same language job resolved against one surface's specific rules, and those rules are documented, so this is the surface where the work is most checkable. When a rival outranks you on an identical product at a comparable bid, the difference is rarely the bid. It is the title string and the eleven fields behind it. Everything in this section applies to Standard Shopping and Performance Max alike, and the UK spelling google shopping feed optimisation refers to the same work. ### How does Google read a Shopping feed, and in what order? Merchant Center resolves a series of questions about your product, and each one can eliminate the item before the next is asked. Bid is not among the early ones. : Ingestion and validation. Items with missing or malformed required attributes are disapproved, and a disapproved item enters no auction at any bid. : Identity resolution. gtin, mpn, brand and identifier_exists resolve your offer against Google's catalog; item_group_id collapses twelve colorways into one product with twelve options. (Classification. google_product_category places the item in Google's taxonomy) one value, numeric ID or full path, assigned by Google if left blank. product_type is your own taxonomy, read as a merchant signal. : Query matching. title is what a query is matched against first, with product_type and google_product_category read alongside it. description, color, size, material, pattern, product_highlight and product_detail refine that match rather than establish it. : Auction and ranking. Bid, landing-page experience, price competitiveness, data freshness. : Segmentation. custom_label_0 through custom_label_4, read by your campaign structure and nothing else. Ingestion, validation and eligibility all resolve before a bid is looked at. ### What is the correct title format for Google Shopping? Google publishes a recommended attribute order per category, and it is not the order most catalogs use. Front-load what a shopper types: the shopping unit truncates long titles, so the identifying words have to come first. The matcher reads the whole string either way. : Apparel and accessories: Brand + Gender + Product Type + Attributes (color, size, material) : Consumables: Brand + Product Type + Attributes (weight, count) : Hard goods and electronics: Brand + Attribute + Product Type + Model number : Books: Title + Type + Format (hardcover, paperback, ebook) + Author : Seasonal and occasion: Occasion + Product Type + Attributes The hard behaviors: title accepts 150 characters; ALL CAPS is a policy violation; promotional text ("Free Shipping", "Sale") is disallowed; phone numbers, URLs and filler like "high quality" burn the characters that render. From an apparel catalog, anonymized: : Before: W-TOP-LIN-2450 Midnight Oasis : After: \[Brand\] Women's Linen Wrap Blouse - Navy Short Sleeve - Workwear - Size 8 "Midnight Oasis" is a real color name from a real PIM. No shopper has ever typed it, and nothing in the Merchant Center tells you so. ### Which twelve feed attributes decide whether you are eligible at all? Twelve fields decide whether Google can validate, identify, classify and match your product. The rest of the specification keeps you compliant; these decide whether you enter the auction at all. Every field below, with what Google does with it and the mistake that most often costs impressions; the accepted values and character limits for each are set out in Google's product data specification (https://support.google.com/merchants/answer/7052112). Attribute | What Google does with it | The mistake that costs impressions title | Matched against first; accepts 150 characters, and the unit truncates the long ones | SKU codes and invented color names at the front description | Refines the match; read closely by AI surfaces | Brand copy with no attributes google_product_category | Places the item in Google's taxonomy | Left blank, so Google guesses product_type | Merchant classification; segments PMax listing groups | Mirrors the buying team, not the shopper brand | Identity; required except books, films, music | "Generic" on branded stock gtin | Matches your offer to Google's catalog, check digit and all | One parent GTIN across every variant mpn + identifier_exists | Identity where no GTIN exists | identifier_exists: no as a shortcut item_group_id | Groups variants into one product with options | Unique per variant, fragmenting the group color, size, material, pattern | Variant definition and attribute matching | Internal color names, no color family product_highlight | Feeds the detail panel; read by AI surfaces | Restating the title instead of adding facts product_detail | Structured specs a machine can parse | Specs buried as prose in the description custom_label_0 to custom_label_4 | Campaign segmentation and bidding only | Believing they influence matching Two rows repay a second read. Custom labels are treated as a lever the matcher never sees. product_highlight is the field most often skipped, and one of the few places to state a benefit in shopper language. ### What did Google add to the Merchant Center in 2026? Google added the fields in two waves, and the second one is the one that lands in your feed. On 11 January 2026 Google announced the Universal Commerce Protocol with more than twenty backers, the Business Agent (a brand-voice sales associate inside Search), and Direct Offers in AI Mode. That wave was protocol and placement. At Google Marketing Live on 20 May 2026 Google shipped the product-data half: six conversational attributes in Merchant Center, rolling out globally, alongside AI Performance Insights reporting. The six fields are: Field | What it carries question_and_answer | The questions shoppers actually ask, answered in their phrasing document_link | Manuals, size guides, spec sheets, care documents related_product | What this item pairs with, fits, or replaces item_group_title | A clean name for the variant family variant_option | The option values within that family popularity_rank | Relative popularity within your own catalog Three things about these fields are worth stating plainly, because most coverage skips them. They are optional. They do not affect product approval, so testing them carries no risk to your existing listings. And they are still unfilled in most catalogs, which is precisely why they are worth filling now. They are submitted through a supplemental feed or the Merchant API, alongside your main feed rather than inside it. Note what Google shipped and what it did not. It built the protocol, the agent, the fields and the reporting. It did not supply the sentences that go inside them. ### Does feed quality matter more in Performance Max or Standard Shopping? It matters in both, but in Performance Max the feed is the targeting. There are no campaign keywords to compensate, so title, product_type and the attribute fields do a keyword list's old job. Listing groups segment on google_product_category, product_type, brand, condition, item ID and custom labels, which makes product_type load-bearing twice: for matching, and for how you slice budget. Search-term visibility is limited, so a weak title fails quietly. Keep a small Standard Shopping campaign as a diagnostic even when Performance Max carries the budget. It still returns search terms, and search terms tell you what vocabulary Google matches your catalog against. ### Which disapproval traps quietly cost impressions? Most lost Shopping impressions are not a suspended account. They are item-level disapprovals sitting in the Products diagnostics tab, unopened since the last feed migration. 1. Price, currency or tax mismatch between feed and landing page at crawl time, including prices that only appear after a cookie banner or region selector. 2. Availability mismatch, usually a stale feed against live inventory. 3. Invalid, duplicated or check-digit-failing gtin values, most often after a supplier import. 4. Promotional overlays, watermarks or borders in image_link, disallowed however good the image. 5. Missing or contradictory returns and shipping policy pages, read on the site, not in the feed. Left alone, these escalate to an account-level notice. If you are already there, work our misrepresentation checklist first, and appeal only once the remediation is complete rather than while it is still in progress. ## What does "optimized" mean on each surface? There is no single definition. Each surface rewards a different property of the same record, which is why one rewrite can move four channels while a surface-specific tool can only ever move one. ### Google Shopping and Performance Max Optimized means the title and attributes overlap with the query at the token level, and the category is precise enough that the auction benchmarks you against the right products. Title construction order and what PMax does with product_type are in the Google Shopping section above. ### Meta Advantage+ Optimized means the catalog carries enough descriptive breadth to build a usable interest signal, and the variant structure does not fragment it. Meta is not matching a typed query, so the value sits in descriptor richness and clean grouping rather than exact-match tokens. ### Onsite search Optimized means the attribute values match the words in your own search box. Zero-result queries are the cheapest vocabulary research available and almost nobody mines them: a literal list of the words your catalog cannot answer. ### AI assistants and AI Mode Optimized means the record can satisfy a constrained request, a specific fabric or fit or occasion, under a price. Assistants read the product structured data on the page as well as the feed. A December 2025 Semrush survey of 1,030 US shoppers who had tried AI tools found that 22% had bought a product directly inside an AI tool, and 50% had made a purchase after researching in one. People discover in the assistant and often buy on the site, which means the record has to earn the recommendation before the click exists. ## How do you prove product feed optimization actually worked? You prove it with a control, and three designs survive scrutiny: a matched-spend A/B test, a geo or audience holdout, and difference-in-differences against a comparable untouched segment. Platform-reported lift is not one of them, because every channel counts shoppers who are already coming and the totals end up larger than the orders you shipped. Here is the anatomy of the number at the top of this page. A catalog was split into a treatment group and an untouched control. Spend was held constant across both, which is what makes the result legible to a finance team, because it removes the obvious alternative explanation. The only variable changed was the product-language record: titles, attributes, descriptors, taxonomy. The window was 28 days, long enough to clear purchase-cycle noise and short enough that catalog churn did not contaminate the groups. In a matched-spend A/B test with a 28-day holdout, the rewritten product-language input produced a 28% increase in Google Shopping revenue against the untouched control. The same design has been run on the other surfaces. On Meta Advantage+, a holdout test cross-validated with Meta's own Conversion Lift measured a 21.4% improvement in ROAS. On owned search, a statistically significant A/B test measured a 28.3% increase in onsite revenue. These sit inside a body of more than 1,000 controlled tests run before the benchmark was published, which is the only reason they are quotable. Here is what they do not prove. They are not a forecast for your catalog, they do not isolate which of the eight steps carried the gain, and they say nothing about categories with thin descriptive vocabulary. A result you cannot interrogate is a slide, not a proof. ## Why does feed optimization decay? Because a catalog is a population that turns over, not a project you finish once. Every week new SKUs arrive carrying the same PIM defaults, the last pass cleaned up, seasonal drops land with empty attributes, suppliers change their descriptors, and the surfaces change what they ask for. That last one is not hypothetical. Merchant Center gained six conversational attributes in May 2026, so a catalog optimized in April was, by definition, blank in fields that did not exist when the work was done. Shopper vocabulary moves too, and it does not send a release note. So the honest way to think about a feed project is half-life. Ship the rewrite, and the share of your catalog carrying optimized language starts falling the next day, at roughly the rate the catalog turns over. A brand refreshing a third of its assortment a season watches the old vocabulary creep back with every new SKU that lands, with a clean error report the whole way down. That makes continuous work a requirement of the arithmetic, not a line item someone invented to bill you twice. ## Would it survive a finance review? That is the compression of the whole evaluation, and it is four questions long. Take them to any vendor, including this one. : Does it fix the input, or the layer above it? Bid management, budget pacing and creative all operate downstream of the product-language record. A tool that cannot change what the product says about itself is redistributing a fixed input. : Does it run continuously, or is it a project? Ask what happens to the 4,000 SKUs landing next quarter. A one-time rewrite is a depreciating asset with no line item for the depreciation. : Does it cover the surfaces that pay this quarter? Paid Shopping, PMax and Advantage+ move revenue now. The assistant surfaces are real and growing, and they are not yet the P&L. A roadmap that serves only the interesting surface is a 2028 budget wearing 2026 clothes. : Can it prove lift against a control? Not attribution. Not a dashboard grading its own homework. A holdout, a named window, and a number that stays the same size when a CFO asks how it was measured. Start with the first question. Every other question in feed optimization sits downstream of whether anyone is fixing the words the product uses to describe itself, and every answer is worth exactly what its method can survive. Next: the field-level reference in product attributes, and the proof design in does feed optimization increase sales. ## FAQ **What is product feed optimization?** Product feed optimization is the work of improving the product-language record (titles, attributes, descriptions and taxonomy) that Google Shopping, Meta Advantage+, onsite search and AI assistants read before deciding whether to show a product. Most work published under that name fixes formatting errors, while the revenue sits in the vocabulary layer, because feeds are scored against how shoppers describe products rather than how a catalog was built. **How long does feed optimization take to show up in revenue?** Syntax fixes clear disapproval within a crawl cycle, but language changes need a window long enough to clear normal purchase-cycle noise, which is why 28 days is a common floor. Anything reported inside a few days is measuring volatility rather than lift. **What is the correct title format for Google Shopping feed optimization?** Google publishes a recommended attribute order per category, and for apparel it is brand, then gender, then product type, then attributes such as color, size and material. The field accepts 150 characters and the shopping unit truncates long titles, so the identifying words a shopper would type have to sit at the front of the string. **Which field in a product feed matters most?** The title carries the most weight on query-matched surfaces because it is scored first against what a shopper typed, and it is also the field most often written in internal style. The highest-value pass is the title plus the filterable attributes (material, pattern, size_system, age_group) that decide whether a product survives a refined search. **Does feed optimization work on Meta Advantage+ as well as Google Shopping?** Yes, though the mechanism differs: Google matches typed queries to titles and attributes, while Meta uses descriptive breadth and clean variant structure to build its targeting signal. On Meta Advantage+, a holdout test cross-validated with Meta's own Conversion Lift measured a 21.4% improvement in ROAS. **Are the 2026 Merchant Center conversational attributes required?** No, they are optional and products without them still serve normally in Shopping. They are the six fields Google introduced at Marketing Live on 20 May 2026, and they are what AI Mode, Gemini and the Business Agent read when a shopper asks a question rather than types a keyword. **How do I prove that feed optimization caused the lift?** Use a control: hold spend constant, change only the product-language record for the treatment group, and compare against an untouched holdout over a named window. Platform-reported lift will not settle it, because each channel counts shoppers who were already arriving and the combined totals exceed the orders actually shipped. **Do I need to redo feed optimization after the first pass?** Yes, because a catalog is a population rather than a project: new SKUs arrive with the original defaults, seasonal assortments land with empty attributes, and platforms add fields that did not exist last time. Merchant Center gained six conversational attributes in May 2026, which left every previously optimized catalog blank in those fields overnight. --- # Product taxonomy: where a product sits, and which queries it is eligible for https://www.lily.ai/blog/product-taxonomy 2026-08-25 A product taxonomy decides which queries your catalog is eligible for. Learn to map it, structure it and maintain it across search, Shopping and AI. Ask three merchandisers where a fleece-lined flannel shirt jacket belongs and you get three incompatible answers. A product taxonomy is the hierarchy that determines where a product sits in a catalog, and it decides which queries the product is eligible to appear for on every discovery surface. Most retail taxonomies are organized around internal structure (buying teams, warehouses, margin bands) rather than around how shoppers describe what they want, which is why products that clearly match a query are frequently not retrieved for it. Google's product taxonomy is a mapping requirement, not a strategy; the strategic layer is whether your own hierarchy mirrors shopper intent. Taxonomy gets filed as an internal-organization problem. It is a discovery problem. ## What is a product taxonomy, and how is it different from an ontology or a category tree? A product taxonomy is a hierarchical classification in which every product has one primary home, reached by a single path of parent and child nodes: Apparel > Outerwear > Coats & Jackets > Puffer Jackets. Five neighbouring concepts get conflated with it. An ontology is a model of how product concepts relate, not only how they nest. A taxonomy says a puffer jacket sits under outerwear; an ontology says it is down-filled, worn as an outer layer, warmer than a shirt jacket. Hierarchy is one relationship; an ontology carries many. A category tree is one rendered instance of a taxonomy, usually the customer-facing one. Navigation is a presentation layer: it hides nodes that exist and exposes landing pages that exist nowhere in the tree. Google product category is a fixed, published list Merchant Center classifies items against; you map to it, you do not design it. And product attributes are the properties of one item: fabric, rise, occasion. Attributes describe the product. Taxonomy places it. The terms below are used interchangeably in most organizations and are not interchangeable. Term | In one sentence | What it governs Product taxonomy | The hierarchy giving every item one primary place. | Eligibility, reporting, assortment Ontology | A model of how product concepts relate. | Synonymy, substitution, intent matching Category tree | One rendered instance of a taxonomy. | What a browse path looks like Navigation | The clickable paths shown to a shopper. | Findability by clicking Google product category | A fixed published list Merchant Center classifies against. | Ad and free-listing classification Attributes | The descriptive properties of one product. | Filtering, matching, retrieval detail ## Taxonomy vs categorization vs attribution: what is the difference? A product taxonomy is the structure, product categorization places an item into it, and product attribution describes the item once placed. Structure, placement, description, each failing differently. Attribution fails visibly: the item appears wrong in a filter, or not at all because the field is blank. Categorization fails silently: a vendor's category string is inherited and the item lands in a node nobody browses. Taxonomy fails structurally, which is the worst, because a bad node is wrong for every item inside it until somebody redesigns the tree. It is why product data enrichment plateaus when nobody questions the hierarchy. ## Why does every retailer run five taxonomies at once? Every retailer already runs several product taxonomies at once, and they are not the same tree: the internal merchandising hierarchy, the site's browse structure, Google's category list, each marketplace's required node, and, new since generative search, the implicit taxonomy an AI assistant builds from the language in your record. The work is mapping between them, then keeping the mapping alive. All five exist in most catalogs today. Who owns each, what reads it, what breaks when it is wrong. Which taxonomy | Who owns it | What reads it | What breaks when it is wrong Internal merchandising hierarchy | Buying and planning | ERP, PIM, sell-through reporting | Planning drifts from what shoppers buy Site browse taxonomy | Ecommerce merchandising | Browse, category landing pages, facets | Category pages go thin; the product cannot be clicked to Google product category | Feed and channel ops | Merchant Center, Shopping, Performance Max | Items are disapproved, mis-bid, or served wrong queries Marketplace taxonomy | Marketplace / channel team | Marketplace browse, search, attribute rules | Listings are rejected, or land in a node with no demand Implicit AI taxonomy | Nobody, currently | LLM retrieval, AI Mode, in-chat shopping | The product is never retrieved, and no dashboard reports it The last row changes the job. The first four are yours to author or map. The fifth is inferred from your product language by a system you cannot query. ## Why do internal hierarchies lose queries? Internal hierarchies lose queries because they encode how the business is organized rather than how demand is expressed. A tree built around buying teams, warehouse zones or margin bands is a good management artifact and a poor retrieval structure. The pattern is easy to spot. Dresses split across two top-level nodes because two buyers own them. "Contemporary," a price tier nobody types. A "Basics" node defined by a vendor agreement. Meanwhile occasion, fabric, fit and warmth live nowhere in the hierarchy, or only as facets that never generate an indexable page. That is criterion (a) in physical form: the problem sits in the input layer, not in the bidding or the dashboard above it. No spend fixes a product that was never eligible. ## How deep should a product taxonomy go? A product taxonomy should be deep enough that every leaf node maps to a distinct thing a shopper asks for, and no deeper. Depth buys precision and costs maintenance; breadth buys coverage and costs relevance. Too shallow, and a leaf holds thousands of items sharing nothing but a department, so ranking inside it is arbitrary. Too deep, and leaves hold a handful of items each: thin pages, unstable reporting, rot. A node that exists only because one buyer wanted a report is a filter wearing a category's clothes. Depth follows demand asymmetrically. Running shoes and foundation shades earn depth; categories where shoppers ask loosely earn breadth plus strong attribution. ## Where do product taxonomies break? Product taxonomies break at the seams: seasonal and occasion categories, cross-category products, gifting, and new product types. Each is a case where the item has one home in the hierarchy and several in a shopper's head. Seasonal nodes are time-bound and the tree is not; a "Holiday" node built in October is still there in March, ranking for nothing. Gifting breaks because "gifts for a new dad under fifty dollars" is defined by recipient, price and occasion, which are attributes rather than places. New product types break because the taxonomy predates the product. ### Worked example: where does a fleece-lined flannel shirt jacket belong? It legitimately belongs in three places. By construction it is a shirt. By use it is outerwear. By occasion it is a transitional-weather layer shoppers name with a word, "shacket," that appears in no formal taxonomy. Resolve it without cloning the SKU: 1. Pick one canonical node and record why. Apparel > Outerwear > Shirt Jackets. Use beats construction, because retrieval follows how people shop. 2. Give the internal hierarchy its own field. Buying still needs it under Wovens/Shirts. A reporting attribute, not a second parent. 3. Express every other home as an attribute. layer: mid-layer, lining: sherpa, fabric: cotton flannel, warmth: light, season: transitional. These drive facets and collections without fracturing the tree. 4. Map google_product_category deliberately. Apparel & Accessories > Clothing > Outerwear > Coats & Jackets, not Shirts & Tops. Write down why; it will be re-litigated otherwise. 5. Carry the vocabulary in the language record. "Shirt jacket," "shacket," "flannel overshirt." A node cannot hold synonyms. The product-language record can, and it is what feeds every discovery surface. One home in the tree, many homes in the language. ## How does taxonomy decide what an AI assistant retrieves? Taxonomy decides AI retrieval by deciding what your product is grouped with in the only structure an assistant can see: the language of the record itself. The assistant builds an implicit taxonomy from the words attached to each item, then retrieves against a shopper's phrasing. So the hierarchy matters only insofar as it surfaces in language. An item titled "Style 4471 Woven Top" has no group; it is not in the consideration set for "casual layering piece for autumn" at all. Google's May 2026 Merchant Center release added six conversational attributes because platforms need this vocabulary and cannot generate it: the field exists, the value has to come from the retailer. In a statistically significant A/B test, rewriting the product-language input produced a 28.3% increase in onsite revenue against the untouched control. That test moved the language, not the tree, which is the honest boundary of what it proves. ## How do you stop a product taxonomy from decaying? A product taxonomy decays because the catalog churns underneath it, so the only thing that stops decay is a maintenance loop running on the catalog's clock, not a project schedule. New product types arrive, vendors change category strings, seasonal nodes outlive their season, and a mapping that was correct when it was written goes stale without anyone editing it. Governance that holds has five parts: : A named owner with authority to add and retire nodes. Not a committee. : A written rule for creating a node, and one for retiring it. Trees rot from unrestricted creation more often than from bad design. : A review cadence tied to the buying calendar, not the content calendar. : Four counts monitored continuously: items in a catch-all node, nodes too thin to rank, items mapped to a parent Google category rather than a leaf, and vendor-supplied categories never reviewed. : A change log, so the next person inherits the reasoning. This is criterion (b), and it is the one taxonomy projects skip. A one-time re-taxonomy starts depreciating the day the next drop lands; a quarterly clean-up is a slower version of the same mistake. ## What to do this quarter Pull the four counts above. Then take your twenty highest-revenue items and ask: does the record carry the words a shopper would use, or only the words your buying team uses? If it is the second, the tree is not your first problem. Then hold what you do next to the standard that matters: one category, one changed input, one control, and a result that survives a finance review. This piece sits alongside product attributes, product data enrichment and product feed optimization. ## FAQ **What is a product taxonomy in ecommerce?** A product taxonomy is the hierarchical classification that gives every item in a catalog one primary place, reached through a single path of parent and child categories. That hierarchy determines which browse pages a product appears on and which queries it is eligible for across search, shopping and AI assistants. **What is the difference between a product taxonomy and a category tree?** A product taxonomy is the classification logic for the whole catalog, while a category tree is one rendered instance of that logic, usually the version shoppers browse. One taxonomy can produce several category trees, and a browse tree often hides or reorders nodes without changing the taxonomy underneath. **What is Google's product taxonomy and is it required?** Google's product taxonomy is a fixed, published list of category nodes that Merchant Center uses to classify items, and retailers map their own categories onto it rather than designing it. Mapping is required for reliable classification in Shopping ads and free listings, but it is table stakes rather than strategy, because every competitor maps to the same list. **How many levels should a product taxonomy have?** A product taxonomy should have enough levels that each leaf node corresponds to a distinct thing shoppers ask for, which in practice is usually three to five levels deep. Categories where shoppers ask precisely justify more depth, while loosely described categories are better served by fewer levels plus strong attribute coverage. **Can a product belong to more than one category?** A product should have one primary node in the hierarchy and express every other legitimate home as an attribute rather than a second parent. Duplicating an item across parents fractures reporting and inventory, whereas attributes such as occasion, layer and season let the same item surface in many collections and filtered results. **How often should a product taxonomy be reviewed?** A product taxonomy should be reviewed on the catalog's cadence rather than a fixed annual schedule, which means every time a new season, product type or vendor category string enters the catalog. Retailers with high churn need continuous monitoring of catch-all nodes and unmapped items, because a taxonomy accurate last quarter is already drifting. --- # The Role of Product Content in AI Search and Discovery https://www.lily.ai/blog/the-role-of-product-content-in-ai-search-and-discovery 2025-12-11 Product discovery is experiencing its biggest shift since digital commerce began. This report gives retailers a framework for structuring product content so generative engines can surface it. [image: The Role of Product Content in AI Search and Discovery] Product discovery is experiencing its biggest shift since digital commerce began. This report gives retailers a clear framework for structuring product content so generative engines can reliably understand, match, and surface their products across modern search and shopping environments. --- # The Readiness Gap: Why 70% of Shoppers Are Ahead of Retailers in AI Commerce https://www.lily.ai/blog/the-readiness-gap-why-70-of-shoppers-are-ahead-of-retailers-in-ai-commerce 2025-12-11 Shoppers are adopting AI-powered shopping faster than retailers can adapt. This report reveals how consumers use AI to search, compare, and buy, and what brands must do to stay visible. Discover why shoppers are adopting AI-powered shopping faster than retailers can adapt. This new report reveals how consumers are using AI to search, compare, and buy, and what brands must do now to stay visible in the age of agentic commerce. --- # 5 Ways Smarter Product Content Can Boost Your Google Ads ROI This Holiday Season https://www.lily.ai/blog/5-ways-smarter-product-content-can-boost-your-google-ads-roi-this-holiday-season 2025-12-02 Holiday 2025 is shaping up to be one of the most competitive seasons on Google, where performance depends less on keyword strategy than on how well Google can understand what you sell. [image: 5 Ways Smarter Product Content Can Boost Your Google Ads ROI This Holiday Season] Holiday 2025 is shaping up to be one of the most competitive seasons on Google. With AI now interpreting everything from product attributes to schema structure, performance depends less on keyword strategy and more on how well Google can understand what you sell. Its systems evaluate product data much like shoppers do, looking for clarity, context, and intent across titles, attributes, and descriptions. Retailers and brands that enrich their feeds with more complete, consumer-friendly product content are already seeing improvements in Quality Score, ad rank, and CPC efficiency. As Google’s models continue to evolve, your product data does more than power discovery. It shapes how your catalog is interpreted, matched to real holiday intent, and ultimately surfaced across Google Shopping, Search and all Google surfaces. It’s not too late to make your product content work harder for you this holiday season. ### 1. Start with Shopper Intent, Not Keywords Holiday searches are emotional and descriptive, e.g. “festive (yet appropriate!) dress for an evening office party.” Lily AI decodes this natural language expression into relevant, structured product data so your ads align with how people actually search. Adding intent-rich attributes (“soft cashmere cardigan in winter white”) will boost quality scores, ad rank and most importantly, click-through and conversion. How Lily AI helps: Lily AI continuously translates real shopper language into structured, AI-ready product content so your Google Ads automatically stay aligned with live consumer intent and perform better across every campaign. ### 2. Optimize Your Data Feeds Specifically for Google’s Requirements As Google’s algorithms and AI agents evolve, product feeds have become the backbone of ad performance. Clean, structured, and continuously updated data determines whether your products appear at all. How Lily AI helps: Lily automates optimized schema and attribute syncing directly within Google Merchant Center, ensuring your product data stays accurate, fresh, and fully optimized for how Google’s AI interprets and ranks retail content. ### 3. Refresh Content Continuously, Tuning for Performance Generative AI search and discovery platforms reward freshness, not just of pricing or availability, but of language, context, and relevance. As consumer behavior and search prompts shift by the hour, static content risks losing visibility and performance momentum. How Lily AI helps: Lily continuously updates an array of product content to reflect live performance signals, seasonality, and trending search and prompt language. Lily also adapts product language based on analysis of consumer engagement data, helping your catalog “learn” what’s working as the season unfolds. ### 4. Bridge Human Language and Machine Logic Today’s AI-driven search rewards brands that speak fluently to two very different audiences: humans seeking inspiration and algorithms parsing intent. The language of commerce is now conversational, contextual, and data-driven so be sure every product is optimized to perform for both people and AI. How Lily AI helps: Lily structures product language that’s emotionally resonant for shoppers, semantically rich for Google’s AI, and precisely optimized for every Google surface. ### 5. Feed AI Systems Clean, Connected Data In an AI-first retail landscape, messy or inconsistent product data is the fastest path to invisibility. Disjointed taxonomies and fragmented feeds prevent algorithms from recognizing your products or showing them in the right context. How Lily AI helps: Lily unifies your product attributes, metadata, and schema into a connected structure, ensuring Google’s AI understands exactly what each SKU represents and when it’s most relevant to surface. The secret to winning this holiday season? Product content that speaks like your customers and thinks like AI. Product Content Optimization (PCO) ensures your catalog is understood accurately and surfaced intelligently, so your Google Ads perform at their best when competition is at its peak. --- # Product Discovery Disrupted: From SEO to GEO and Beyond https://www.lily.ai/blog/product-discovery-disrupted-from-seo-to-geo-and-beyond 2025-06-10 See why AI disrupted search and shopping for the better, and how to succeed in the new world of AI-powered search and shopping across paid, organic, and owned on-site. [image: Product Discovery Disrupted: From SEO to GEO and Beyond - From Lily AI] See why AI disrupted search and shopping for the better, and how to succeed in the new world of AI-powered search and shopping across paid, organic, and "owned" on-site. Download our new playbook to see: - The technological and consumer trends shaping search and shopping today, from SEO to GEO and beyond - The language gaps that exist between how merchants and marketers describe products, and how consumers and machines discover products - 3 steps for getting started with Product Content Optimization (PCO) to bridge the gaps - Case studies of retailers and brands that boosted product visibility, discoverability, and sales with PCO AI is more than hype and headlines; it is visibly reshaping shopping in ways that consumers are fast adopting. Download the Product Discovery Disrupted: From SEO to GEO and Beyond playbook to learn the new strategies for SEO, GEO, SEM, on-site, in social, and beyond. ## What's Inside ### State of AI in Shopping Generative and Agentic AI innovations are creating disruption almost daily. ### Consumer Trend Shockwaves Technology has advanced, and consumers have already adapted. ### Decoding Product Languages How brands describe products is not typically how ordinary people shop. ### Product Content Optimization (PCO) The critical missing layer in retail’s tech stack. ### Getting Started with PCO 3 steps to maximize discovery in the AI Era. ### PCO Case Studies The impact of Lily AI Product Content Optimization for brands and retailers. --- # What is GEO vs SEO? Retail FAQs https://www.lily.ai/blog/geo-vs-seo-for-retail-and-shopping-faqs 2025-05-20 Ask 10 CMOs whether they should be concerned about GEO and you will get 10 different answers. Here are the most common questions about generative engine optimization, unpacked. [image: Example of AEO with a prompt on Perplexity displaying a shopping image of shoes to choose from.] ## Key Takeaways: - Defining GEO today and why it’s both similar and different from SEO. - GEO strategies require enriched structured data and natural language content. - Winners in the GEO world benefit from increased brand and product visibility as well as clicks. If you ask 10 CMOs whether they should be concerned about GEO, you’ll receive 10 different answers. - Pragmatists are concerned about zero-click search and 15-25% drops in organic traffic from Google. - Optimists see the opportunity to organically reach new customers in more channels. - Skeptics want to first agree on the same definition and ensure you’re not confusing GEO with SEO. The skeptics aren’t wrong, either! Even the right acronym to use isn’t settled among marketers yet. A Digiday WTF are GEO and AEO explained: “This is such a new area that currently there is no common taxonomy. So agencies, publishers, marketers and SEO specialists have adopted a bunch of different acronyms to describe the same trend.” ## Demystifying GEO: FAQs “Generative Engine Optimization” is certainly a hot topic in retail, so let’s unpack some common questions: 1. What is GEO? 2. Are consumers using GEO for shopping? 3. How is GEO different than SEO? 4. How should brands and retailers best approach GEO? ### What is GEO? First, let’s define GEO: an emerging new discipline that enhances digital content to be found and favored by AI-powered search engines. The retailer’s goal is similar to Search Engine Optimization, which involves organically increasing product visibility and driving more traffic to an e-commerce website. GEO is similar to many other related yet distinct terms, such as: – AEO (Answer Engine Optimization) – GSO (Generative Search Optimization) – AISO (AI Search Optimization) – AIO (AI Overviews Optimization) – LLMO (Large Language Model Optimization) Another term often related to GEO is the idea of “zero-click” answers or search. Similar to featured snippets on Google, zero-click is when a satisfying answer doesn’t require clicking to a website for further information. In the context of retail and shopping, the GEO experience feels similar to Google in many ways. When someone searches for “Best breathable, slim-fit pullover with pockets for all-day hikes?” or “Recommended durable, pet and kid friendly sectional for playroom,” those searches return product image carousels and listings with helpful, natural language explanations behind each recommendation. With OpenAI launching a shopping experience inside of ChatGPT in April 2025, and Perplexity launching a similar shopping experience with “Buy with Pro” in November 2024, not only are these GEO experiences providing product recommendations, but purchases can actually be made directly within the platform. ### Are Consumers Using GEO for Shopping? On average, about 4/10 U.S. consumers are using Generative AI and Answer Engines for Shopping, or are relying on those results: - 41% have considered AI-powered search when shopping (eMarketer) - 4/10 U.S. consumers used AI-powered search for shopping, (Lily AI) Lily AI 2025 Consumer Shopping Survey What does this mean for retailers? Answer Engines open up a new channel to potentially reach more new customers, but it’s also a missed opportunity if you’re not optimizing to gain share of voice and missing out on brand and product mentions. ### How is GEO Different Than SEO? Fortunately, many of the same strategies that are the lifeblood of SEO are also universally beneficial for GEO. Let’s start with SEO. As any digital marketer or agency will tell you, SEO is dynamic as Google algorithms continuously evolve. To stay one step ahead, holistic SEO traditional strategies are built upon three core pillars: 1. Content SEO: This includes both consumer-facing product copy and non-consumer-facing metadata and schema markup for products. 2. Off-page SEO: Think backlink strategies, online reviews, and brand and social media mentions. 3. Technical SEO: Core web vitals, page speed, mobile-friendliness, and site architecture, etc While all of these areas are important, in GEO, the content aspect takes center stage. Specifically, the “machine speak” of non-consumer-facing data on your landing pages and Product Display Pages (PDPs). ### How Should Brands and Retailers Best Approach GEO? Investing in product content has tremendous value for both GEO and SEO. Specifically, product content should be optimized for “machine speak,” including enhanced schema markup for products and meta titles and descriptions. Yet content enrichment is not simply adding a few extra keywords or overdoing it with keyword stuffing. Rather, Gen AI and LLMs prefer content and metadata that demonstrate a much higher bar of both quality and completeness, which is both purposeful and contextual. For example: - Maximum Attribute Coverage: Go beyond the basics, such as detail nuanced fit (e.g., “relaxed sleeve fit”), specific design elements (“ruched detailing at waist,” “vamp coverage” for shoes), or color families (cool vs warm or neutral tones). The more attributes, the better, as long as they are accurate and helpful (a YKK zipper is not a selling point). - Description Precision: Clarify ambiguities. Eliminate misleading hypernym and hyponym traps. Is it a “sweatshirt” or a “pullover”? A “face lotion with SPF” or a “Face sunscreen”? This precision prevents misinterpretation by both users and machines. - Natural Language Relevance: Think about how people ask conversational questions. Use words and phrases that reflect natural language, including more subjective descriptors, such as room setting, style, occasion, trend, etc. ## The Universal Language of Product By investing in this level of detail for the product content on your landing pages and PDPs, you’re not just ticking boxes for algorithms. You are: - Enhancing SEO: Improving your chances of ranking for a wider range of specific, high-intent keywords. - Future-proofing for GEO: Positioning your products to be the direct answers in AI-powered search and discovery experiences. - Improving User Experience: Helping customers find exactly what they’re looking for, leading to higher satisfaction and conversion rates. - Streamlining Operations: Well-structured data can often be repurposed for internal analytics, merchandising strategies, and even feed management for advertising platforms. Investing in comprehensive, structured product content for your online catalog (aka your website!) isn’t just beneficial—it’s a universally winning strategy for both traditional search engines and the AI-powered answer engines. In other words, investing in Product Content Optimization for SEO and GEO is the gift that keeps on giving! --- # What is Product Content Optimization? 3 Things to Know https://www.lily.ai/blog/what-is-product-content-optimization-3-things-to-know 2025-04-16 In an AI-dominated retail world, basic product copy and bare minimum product data no longer cut it. Three things every retailer, brand, and agency needs to know about product content optimization. [image: What is Product Content Optimization? 3 Things to Know] In an increasingly AI-dominated retail and e-commerce world, basic product copy and bare minimum product data simply no longer cut it. Both because: 1. High-quality product data is rocket fuel for both traditional and AI-powered search and discovery 2. AI can enrich and optimize the product data itself, at the source Retailers and brands must invest in a product content optimization strategy that can dynamically keep up with today’s consumers and is fluent in what we refer to at Lily as consumer, merchant, marketer, and machine “speak” (we’ll touch more on that later in the post). If it all starts with product content, what is “Product Content” defined as, exactly? – Product content includes the attributes and copy that consumers see, as well as invisible metadata. – Product content optimization refers to the dynamic process of improving and expanding product information, such as adding more details, attributes, specifications, descriptions, highlights, and other labels or tags to a catalog, or more specifically, the product record, product listing, product ad, or product detail page (PDP). Here are the three things every retailer, brand, and digital agency needs to know about product content optimization. ## 1) Product Content Is No Longer Static Gone are the days when product content generation was a one-time exercise when a new product was brought to market. Consumer language is dynamic, so your product content should be too. Consider that: – Seasons change (a spring dress becomes a summer dress) – New trends emerge from micro trends to macro trends – Consumer vocabularies evolve and also vary by region All of this contributes to how today’s consumer describes products differently than even yesterday, as context changes. After all, SEO was never static and thrives on fresh content! Product content optimization today is a dynamic feedback loop that interprets, distills, optimizes, and distributes enriched product content to all destination systems. ### By the Way, Do Product Descriptions Still Matter? Yes! In fact, Lily AI’s 2025 consumer shopping survey showed that 52% of consumers report that short product descriptions are one of the most commonly used content to help with purchase decisions. From the research, 44% also reported using longer product descriptions to inform their shopping decisions. [image: ] ## 2) Product Content is Not Just What Consumers See Consumer-facing product content includes both titles and descriptions but also product images, videos, and reviews, to name a few. All of these help consumers evaluate products. But which of these helps consumers find the products in the first place? To be sold, products first need to be found. That’s where product data comes in, and more specifically, metadata and product data feeds. ### Product Feeds: From Google to Amazon to Meta, you name it, it’s a product data feed world when it comes to online shopping. Product feed platforms, such as ProductsUp, Feedonomics, and DataFeedWatch, enable the seamless distribution of product information across various channels, including marketplaces, search engines, and social platforms. This provides ultimate efficiency and control in delivering engaging product experiences everywhere. Yet, distributing product data is only half the battle; enriching and dynamically optimizing the product data is another story. In product feeds, product titles and descriptions are optimal places to test and customize the integration of new product attributes and descriptors, such as trends, styles, occasions, seasons, and synonyms. Oftentimes, there are also additional fields for the product attributes. These valuable fields are often left blank or incomplete because neither the marketer nor their digital agency has the data to complete them. And who can blame them? With Google alone, it’s hard enough to keep up with product data specs as it is! – Pro-Tip: Lily AI’s GMC integration makes product content optimization in Google a breeze, and our no-code platform seamlessly integrates with feed management systems to optimize product listings and ads everywhere! ### Product Metadata: Metadata encompasses detailed product information—such as attributes, categories, and tags—that enhances search engine visibility and ensures products appear in relevant customer searches. By enriching the product metadata and schema markups on the PDP, brands and retailers can add more structured depth to their product information to boost search engine, answer engine, and Agentic AI discoverability. [image: ] An added bonus? There are two, actually! The same metadata enrichment optimizes: 1) Site Search: Regardless of which site search platform you’re using. From site search leaders Algolia to Bloomreach, to built-in-house systems or off-the-shelf Google site search solutions, all search is optimized with enriched product data. 2) Search Ads: From traditional Shopping Ads to PMax, enriched PDP copy and metadata increase landing page relevance. Landing page relevance improves ad quality scores, which helps advertisers boost ad visibility and cost-effectively drive more impressions and clicks, without increasing media spend! ## 3) Product Content is Not (only) Merchant Speak Product content enrichment refers to the process of improving and expanding product information to make it more complete, accurate, and compelling for consumers, as well as merchants, marketers, and machines. Let’s dive into what that really means here. ### Consumer Speak: Connect with Intent Consumer speak is a dynamic, ever-evolving natural language lexicon influenced by (but not limited to!) pop culture and social media. From catching vibes to vibe coding, the way we all speak changes almost by the day these days! To complicate this further, even how we search varies by platform. On Amazon, you might simply search for “Accent white chair” yet on ChatGPT your prompt is much longer and more complex. [image: ] As new terms emerge and gain traction through social platforms and cultural phenomena, they influence how people express identity and engage with brands. For marketers and merchants, staying attuned to this shifting linguistic landscape is crucial to crafting messages that resonate authentically with their audience. And while both micro and macro trends certainly inspire shopping decisions and motivations, so too does the need to shop by the occasion or need, to help achieve whatever desired aesthetic or functional purpose. This is more than the behavioral language insights gleaned from clickstream or on-site search data. To capture the most holistic view of consumer language today, retailers and brands need to harness both e-commerce data as well as SEO and social media data. ### Merchant Speak: Ensure Accuracy and Consistency Product content optimization starts with merchandising teams since they are critical stakeholders in any product data strategy. Before any product data optimization exercise, the core details, such as item type, features, and color, must first be verified. Every time data changes hands between the manufacturer to wholesaler, brand, retailer, marketplace, and resaler, it’s an opportunity for data loss or degradation. Once product data accuracy is confirmed (an excellent use case for Computer Vision and Natural Language Processing, in particular!), then begins the data enrichment process. This includes: – Filling Gaps: Such as embellishment details or more nuanced fit elements (sleeve fit), that might have been skipped or overlooked. – New Subjective Details: Adding additional details that can be harder to objectively define, such as trends, styles, occasions, or micro-occasions. – Adding Synonyms: Ensuring relevant synonyms are captured for all of these attributes, since back to the previous point above around consumer speak, everyone talks, searches, and shops differently. Think trainers vs tennis shoes vs sneakers, as an example. ### Marketer Speak: Optimize for Discoverability Product content requires meticulous enrichment and dynamic refreshes for optimized visibility across organic search (SEO), paid search (SEM), and e-commerce on-site search. – Natural Language Keywords: Use insights from consumer speak to naturally incorporate relevant keywords (long-tail, question-based) into consumer-facing product copy as well as invisible metadata. – Brand Voice Alignment: While this almost goes without saying, consistency and agreement with brand guidelines are always worth mentioning, especially in the context of consumer-facing content! – Optimized Google Merchant Center (GMC): Accurate, detailed product feeds are vital, especially for platforms like Google Merchant Center (GMC). Whether paid Product Listing Ads or organic Product Listings, high-quality product data is rocket fuel for optimized organic and paid performance. Fortunately, enriching and managing product content and attributes at scale in the Google Merchant Center is easier than ever with Lily AI! ### Machine Speak: Be Found by Google and AI How can brands and retailers communicate more effectively and efficiently with the search engine giant Google, as well as AI-powered search and AI Agents? By harnessing highly detailed, structured, consumer-friendly product content in every digital channel. – Schema Markup for Context: Implement structured data (Schema.org) like Product, Offer, Review, and FAQPage schema. This helps search engines and AI understand context, enabling rich snippets (price, ratings in search) and powering accurate AI-driven answers. – Metadata Unlocks Relevance: Rich metadata—data about your product data—provides essential context for AI. It improves data quality, enables personalization, enhances discoverability, and makes AI outputs more interpretable. AI itself can help extract and manage this metadata. – The Feedless Future: As part of the 2024 launch of Google Merchant Center Next, GMC can now automatically source new products from an e-commerce site through a feedless process called “Found By Google”, in addition to the feed-driven “Provided By You” data source. This means that enriched and optimized PDP product data will only continue to grow in strategic importance! ## Enrich Product Content with Lily AI Effective product content optimization demands fluency in all four languages: consumer, merchant, marketer, and machine. Mastering this multi-faceted approach—keeping content dynamic and valuing both consumer-visible copy and invisible metadata—is crucial for enhancing discoverability, driving greater productivity and new efficiencies, as well as boosting web traffic and sales. And we would know, Lily AI is the leader in AI-powered product content optimization, and we see the impact for our clients every day! Lily AI is transforming eCommerce and advertising by using AI to decode consumers’ true shopping intent and bridge the language and metadata gap between how retailers describe their products and how consumers express what they want. Leveraging a suite of advanced AI technologies fueled by high-quality, human-verified proprietary data, Lily optimizes product content, enabling retailers to understand complex consumer search behaviors, improve product attributes, titles, and descriptions, and personalize shopping experiences, resulting in operational efficiency and increased sales. Want to learn more or schedule a demo? Let’s talk! --- # The Search for Smarter Shopping: The Power of Product Content https://www.lily.ai/blog/the-search-for-smarter-shopping-the-power-of-product-content 2025-02-19 Findings from a consumer shopping survey conducted by Lily AI in January 2025, on the heels of the 2024 holiday season, assessing consumer sentiment toward online search and discovery. [image: The Search for Smarter Shopping: The Power of Product Content] See the hot new findings from a consumer shopping survey conducted by Lily AI in January 2025, conducted on the heels of the 2024 holiday shopping season, to assess consumer sentiment and preferences towards online search and discovery. --- # A-Z Glossary: Key Concepts in AI-Powered Search, Commerce, and Advertising https://www.lily.ai/blog/a-z-glossary-defining-key-concepts-in-ai-powered-search-commerce-and-advertising 2025-01-27 Do you know your product attribution from your measurement attribution? An A-Z glossary of product attributes and content enrichment, AI search and discovery, measurement, and agentic AI. [image: A-Z Glossary: Key Concepts in AI-Powered Search, Commerce, and Advertising] Do you know your product attribution from your measurement attribution? Don’t worry if not; this A-Z glossary has you covered! This glossary is jam-packed and organized into four sections: 1. Product Attributes and Content Enrichment 2. AI Search and Discovery 3. Measurement and Analytics 4. Agentic AI and Machine Learning Feel free to casually scroll or hit that Command+F; and don’t forget to bookmark this page! ## Product Attributes and Content Enrichment ### AI-Powered Product Attribution: An AI-driven process, as opposed to the common human-powered merchant review, of automating the validation of existing product data as well as generating a complete set of additional product attributes. The AI involved typically includes, but is not limited to, Computer Vision and Natural Language Processing. ### Dynamic Attribute Taxonomy: A living library of consumer-oriented attributes and product details that often uses AI to analyze search data to extract up-to-the-minute natural language terms. ### Micro-Trend: A rapidly emerging, short-lived style or phenomenon that gains quick popularity within a niche audience, primarily through platforms like TikTok and Instagram, and may or may not result in a new aesthetic with a hashtag and even a “core” attached, like “#SirenCore”, #MermaidCore’s darker cousin. ### Natural Language Product Attributes: The rich vocabulary that consumers use when describing the product details and characteristics of both past purchases and how they, in their own words, evaluate new items. ### Product Attribution: The process of analyzing product information and images in order to validate information as well as identify missing characteristics and mico-details in order to apply this data to enrich product records, online product listings, and product display pages, as well as a foundational data source of truth to generate additional product content, such as descriptions, highlights, styles, etc. ### Product Attribute: A specific characteristic or feature of a product that describes its properties, functionality, or style. ### Product Content Optimization: An AI-driven, continuous process of enriching and customizing product data across all channels and platforms. It leverages a dynamic feedback loop that includes a brand or retailer’s advertising and e-commerce data to further enhance both consumer-facing and machine-optimized product information with the latest consumer behavior trends and preferences, maximizing discoverability. ### Product Data Enrichment: Product data enrichment refers to the process of improving and expanding product information, such as adding more details, specifications, descriptions, highlights, and other data, to a catalog to make it more complete, accurate, and compelling for customers. ### Product Data Validation: The identification and rectification of inaccurate information in datasets, such as an item category classification for attribute detail. ### Product Tag / Tagging: The application of descriptive labels to online merchandise, including keywords and categories, to boost discoverability and improve search results online. ### Product Taxonomy: A hierarchical classification and descriptor normalization framework for organizing and standardizing merchandise based on characteristics and relationships. ### Special Occasion: There are the traditional occasions for when an item is needed, such as a “black-tie event,” and there are a myriad of special occasions that are less common yet still have distinct and nuanced dress codes, such as a “Coachella Concert Outfit.” ### Taxonomy Alignment: The process of mapping and harmonizing two different sets of product data, such as between a brand and a retailer or a marketplace. ### Zillennials: A “hybrid” micro-generation born between 1994 and 2000, wedged between Millennials and Gen Z with unique cultural and technological experiences. They are only one example of how shopping behaviors and preferences can vary from audience segment to segment and from generation to micro-generation. ## AI Search and Discovery ### Ad Quality Score: In Google Ads, this is a 1-10 rating that measures the effectiveness and relevance of keywords, ads, and landing pages in Google Ads campaigns. The score is calculated based on three key components: expected click-through rate (CTR), ad relevance, and landing page experience, which are evaluated by comparing performance against competitors bidding on the same keyword. A higher Quality Score leads to lower cost-per-click (CPC), better ad positioning, and improved overall campaign performance, making it a critical metric for advertisers to optimize their digital advertising efforts. ### Answer Engine Optimization (AEO): A cutting-edge content strategy that optimizes digital content to provide direct, concise answers to user queries across AI-powered platforms like voice assistants, chatbots, and search engines. Unlike traditional SEO, AEO focuses on structuring content to deliver precise, zero-click answers that can be immediately extracted by AI technologies, prioritizing user intent and conversational language. ### Faceted Search: Also known as Filters and Facets, this is an e-commerce filtering method using characteristics to progressively narrow down large selections that optimize product discovery and rely on product data that is aligned with that e-commerce site’s taxonomy. ### Generative Engine Optimization (GEO): An emerging new discipline of enhancing digital content to be found and favored by AI-powered search engines, with the retailer’s goal being, similar to Search Engine Optimization, organically increasing product visibility and driving more traffic to an e-commerce website. ### PerformanceMax Campaigns: Also known as PMax, this is Google’s AI-powered campaign type that automatically manages and optimizes advertising campaigns and media dollars across multiple Google channels (or “Surfaces”) as well as ad units. ### Product Discovery: A multifaceted process that helps people find and explore new products, typically online, that meet their needs or desires. It encompasses various digital touchpoints and strategies designed to guide shoppers toward items they’re looking for or might be interested in, even if they weren’t explicitly searching for them. ### Product Listings (PLs) vs Product Listing Ads (PLAs): While Google’s Product Listing Ads (PLAs) are now known as Shopping Ads, many in retail still refer to these ad units and organic placements as PLAs and PLs. On Google, the same product information is synced with Google Merchant Center (GMC) to power both PLs and PLAs. ### Retail Media: Also known as Commerce Media, this refers to advertising placed within retailers’ websites, apps, and physical stores, allowing brands to promote their products at or near the point of purchase. The most popular retail media ad unit is the Sponsored Product Listing due to its effectiveness in driving highly profitable, bottom-of-the-funnel sales by appearing prominently in search results and category pages. ### Search Engine Optimization (SEO): The strategic process of optimizing digital content to improve website visibility and ranking on search engine results pages, ultimately driving organic traffic. In the context of retail, it is the process of enriching catalog data and digital product content in order to enhance discoverability and conversion across traditional search engines as well as emerging AI-powered platforms. See AEO and GEO for more specifics on this emerging channel. ### Semantic SEO: The process of optimizing a website and product descriptions to provide content that’s relevant to user intent by understanding the contextual meaning behind a query, rather than matching keywords alone. While semantic search produces superior results to keyword-only matching, it still requires product content enrichment for optimal semantic search performance. ### Shopping Ads (FKA Product Listings): An ad unit on Google that is a highly visible, visual product advertisement that displays relevant products at the top of search engine results. These rely on detailed merchandise information managed in Google Merchant Center in order to optimize ad performance and relevance. ### Sponsored Listings: An ad unit similar to Google Shopping Ads and found in Retail Media that displays sponsored product ads within on-site search results and product listing pages. Like Google Shopping Ads, Sponsored Listings also rely on detailed merchandise information in order to optimize ad performance and relevance. ### Transformer Search: An AI-driven approach that leverages deep learning models to enhance product discovery and search experiences in e-commerce. This technology utilizes transformer architectures to analyze and understand complex relationships between user queries, product attributes, and consumer behavior. While transformer models can work with minimal product data, enriched product attributes significantly enhance model performance. ### Vector Search: A method representing items and queries as mathematical vectors, enabling more nuanced and context-aware results in online shopping environments. Like semantic and transformer search, it produces superior results as compared to keyword-only matching, yet it still requires product data enrichment, given that the quality of vector search results depends on the richness of the product data used to generate the vectors. ### Zero-Click Search: When a user finds the answer to their search query directly, without needing to click through to a website. This trend challenges traditional e-commerce traffic models by reducing website visits while simultaneously offering retailers an opportunity to optimize product visibility through enhanced data presentation and rich search results. Successful adaptation requires retailers to focus on advanced search technologies, semantic search optimization, and creating compelling, information-rich product listings that can capture consumer attention directly within search engine interfaces. ## Measurement and Analytics ### A/B Testing: Also known as split testing, is a method of comparing two versions to determine which performs better in achieving specific goals. In this process, one version (the control) might be shown to one segment while the other version (the variant) is displayed to another segment, allowing for statistical analysis of user behavior to identify the more effective option based on predetermined metrics. ### AI Slop: Also referred to simply as “slop”, this is a term for low-quality media, including writing and images, made using generative artificial intelligence technology. ### Average Revenue Per Visitor (ARPV): A metric calculating the average income generated by each website visitor, useful for assessing overall performance in digital retail. ### Cold Start Forecast: A model of predicting sales for a new product when there is little to no historical sales data available, creating a “cold start” challenge for accurate prediction. This is quite common when launching new items in retail and typically relies on analyzing attribute similarities with existing products. ### Contrafactual DiD: Also known as Counterfactual in DiD, it is an advanced version of the difference-in-difference method incorporating alternative scenarios to estimate causal effects more accurately in e-commerce. ### Difference-in-Difference (DiD): A statistical technique comparing outcome differences between treatment and control groups before and after an intervention in digital retail. ### Fractional Factorial: An experimental design testing a subset of all possible factor combinations, reducing required tests while providing valuable insights for online retailers. ### Full Factorial, Double-Blind: A comprehensive experimental design with both researchers and participants unaware of group assignments to prevent bias in e-commerce studies. ### Interrupted Time Series: An analytical method evaluating the impact of an intervention by comparing data trends before and after a specific point in time for online retail. ### Linear Extrapolation: A prediction technique extending a linear trend beyond the original observation range in data analysis. ### Marketing Attribution: Measurement model that is channel-based in orientation and analyzes multiple touchpoints across campaigns and advertising media. The goal is to identify the impact of each touchpoint by assigning a value to each one that contributed to a desired outcome, typically a conversion or sale. Examples of attribution models include include multi-touch attribution or closed-loop attribution. ### P-Value: Also known as probability value, this is a statistical measure that quantifies the likelihood of obtaining observed results if the null hypothesis is true. ### PDP View Rate: An e-commerce metric that measures the percentage of users who view a product detail page (PDP), providing insights into user engagement and interest in specific products, which are crucial for optimizing conversion rates and improving overall sales performance. High PDP view rates often indicate effective marketing and product discovery, while low rates may signal issues with product visibility or demand. ### Type I & Type II Errors: Statistical mistakes where Type I is an incorrect positive and Type II is an incorrect negative in e-commerce data analysis. ### Zero-Result Search Rate: Also known as the null result rate, it is a metric that measures the percentage of searches that return 0 results on a website or search engine. ## AI Agents and Machine Learning ### Agentic AI: Artificial intelligence (AI) systems designed to autonomously pursue complex goals and execute workflows with limited direct human supervision. It is designed to handle complex, multi-step processes across dynamic environments. ### Classification Models: Machine learning algorithms that sort data points into predefined groups or classes based on their features. These models learn to identify and assign new, unseen data to specific categories by analyzing characteristics from training datasets. ### Computer Vision (CV): An AI field enabling machines to interpret and understand visual information, such as product images and videos. ### Deep Learning (DL): An advanced machine learning subset capable of processing vast amounts of unstructured data for e-commerce applications. ### F-1 Score: A crucial evaluation metric in machine learning, particularly for classification tasks. It combines the concepts of precision and recall into a single score, providing a balanced measure of a model’s performance. ### Human-in-the-Loop: An AI approach incorporating human feedback in the machine learning process, improving accuracy and domain expertise depth. ### Labeled Dataset: This term specifically refers to a collection of data points that have been annotated with meaningful labels or tags that indicate the correct output or category for each data point. Labeled datasets are essential for supervised machine learning tasks, where models learn to make predictions or classifications based on the examples provided in the dataset. ### Large Language Models (LLMs): AI models trained on extensive text data, capable of understanding and generating human-like text for various tasks. ### Machine Learning (ML) A subset of AI that enables computers to learn from data and improve their performance on tasks without being explicitly programmed. ### Natural Language Processing (NLP): Technology enabling machines to comprehend, interpret, and generate human language, facilitating natural interactions in e-commerce platforms. ### Neural Networks: Computing systems inspired by biological brain structures, designed to recognize patterns and learn from large datasets in online retail applications. ### Precision: A critical metric used to evaluate the performance of classification models, particularly in scenarios involving imbalanced datasets. It measures the accuracy of positive predictions made by the model. This metric is essential when the cost of false positives is high, as it indicates how many of the predicted positive cases are actually relevant. ### Recall: A performance metric in AI that measures the model’s ability to identify all relevant instances within a dataset. Specifically, recall quantifies the fraction of relevant instances that were successfully retrieved by the model. Recall is most effective when used in conjunction with precision, typically balanced by the F1 score, which provides a harmonized evaluation of a model’s performance. ### Training Data: Specialized datasets used to train AI models. In retail, retail-specific training data is crucial for developing machine learning algorithms that can accurately generate product attributes and related product data. --- # How AI Connects With Consumers Shopping for the Holidays . . . Wherever They Are! https://www.lily.ai/blog/consumers-shopping-for-the-holidays 2024-08-13 Economic uncertainty and a shorter holiday season make 2024 feel faster-paced and unpredictable. Retailers cannot just remind holiday shoppers they exist; they have to inspire them. [image: How AI Connects Retailers to Consumers Shopping for the Holidays] Economic uncertainty and a shorter holiday season are combining this year, making the 2024 holiday retail season feel faster-paced and a bit unpredictable. It’s critical that brands and retailers don’t just remind holiday shoppers of their existence. As Abhijit Shome, Executive Advisor to Lily AI says, retailers need to inspire consumers to make a purchase. This is especially true in order to survive the year’s biggest shopping season, which will bring on some new retail challenges. We expect that, this year . . . 1. Search engines will still be big. In fact, 64% of people use search engines in their “I-want-to-buy” moment, indicating a significant reliance on them during the shopping process. 2. Social media will be used as a discovery tool. Most consumers are predicted to view social platforms as a way to discover products (59%). 3. Some purchases will be made through social media, though! When using social media for holiday shopping, consumers are pretty evenly split between intentionally using the search function (30%), scrolling until they see something they like (36%), or using both methods (34%) to shop. 4. People still shop both online and in stores. In fact, 79% of shoppers conduct product research online and 37% of those consumers proceed to buy their items in physical stores. All this is to say that social media will provide big opportunities for retailers, especially those hoping to capitalize on the micro- and macro-trends social apps like TikTok have gotten us to love. All the while though, consumers will continue searching for products in more traditional means, like search engines and on-site search. Retailers make the most of all these search methods. With that in mind, here are some ways to connect with holiday shoppers, wherever and however they search. ## How Can AI Strengthen Connections Between Retailers and Consumers Shopping for the Holidays? ### Tip #1: Help people find the perfect gift–wherever they search. Including highly detailed product information within a product listing makes items more discoverable to shoppers . . . no matter where or how they’re searching for products. Remember how 64% of consumers use search engines in their “I-want-to-buy” moment? Here’s where you can use AI to meet them exactly there. One way retailers can do this is by using AI to quickly and thoroughly enhance Google Product Listing Ads, so they are rich with customer language. Lily AI injects product details and attributes that reflect new and evolving trend terms throughout your digital catalog. Its direct integration into Google Merchant Center makes it easier than ever for brands and retailers to enrich their product listing ads with these customer-centric attributes that will surface your products where people search. You can also use AI to reverse-engineer product pages, optimizing SEO and site search, with that same customer language. Your product pages can be a retailer’s ultimate recommendation engine . . . If they are optimized for AI-driven search. So instead of merely listing product features, you should incorporate richer attributes, a greater volume of product details, and multimedia that shows off a product’s features more thoroughly. This will enhance both organic and paid discoverability of products, while making them easier to find on your own site. ### Tip #2: Consumers will search for specific occasions, attributes, AND micro-trends. So capitalize on all of them. Thanks to the influence of social media, trend cycles move faster than ever. And along with the new macro- and micro-trends that crop up all the time on TikTok, Vogue Business predicts that consumers will soon shop for items to wear to micro-occasions. For example, we’ve already seen people search for clothes that reflect the Brat Summer aesthetic. If Vogue Business is correct, they will search for specific occasions to wear their Brat-inspired outfits to, like concerts or nightclubs. This is where we turn to AI to supercharge product listings with customer language. Lily AI’s product attributes help reflect styles, occasions, details, and more, so brands and retailers can get items in front of customers’ eyes as quickly as possible. Its extensive taxonomy of 25k+ attributes for fashion, home, and beauty products includes the following: - Objective Attributes: Including item types, materials - Subjective Attributes: Including styles, occasions - Attribute Synonyms: Such as “glossy” vs. “high shine” - Macro and Micro Trends: Such as Brat, Modern Farmhouse, or Clean Girl - International Language: Including localized terms, translations Injecting these attributes across your products means that you can capitalize on trends as they unfold, connecting with consumers at the peak of trend popularity or seasonality. For example, Revolve does an excellent job with this! Throughout the season, they’ve capitalized on the Brat Summer trend as it boomed, helping shoppers find lime green dresses and Y3K-inspired accessories for their next big nights out. ### Tip #3: Use data to push personalized product recommendations. Meet shoppers in the moment . . . and inspire them to add another item or two to their carts. Recommending highly relevant products to your customers is a critical upsell opportunity that you don’t want to miss out on! This is because: - Personalization increases the likelihood of a prospect purchasing by 75%. - 56% of customers are more likely to return to an e-commerce site that offers personalized product recommendations. - 74% of customers feel frustrated when website content is not personalized, suggesting personalization positively impacts purchase intent, according to our partner Bloomreach. Enriched product data, enhanced by the ultra-specific product taxonomy we previously discussed, presents customers with items that reflect the tastes, trends, preferences, and occasions they are shopping for. It also recommends products that pair well with items they’ve previously shown interest in. For example, if someone added a Coastal Cowgirl sundress to their cart, AI-powered product recommendations will present that consumer with highly-relevant upsell options, from accessories to glam up their look, to cowgirl boots that’ll match its colors and patterns best. ## Connect With More Customers As They’re Shopping for the Holidays There are so many ways to connect with potential customers this holiday season, and AI will supercharge these opportunities! By enhancing your entire customer journey via AI-powered, customer-centric language, retailers will unlock the most effective way to help more people discover their products, no matter how or where they’re searching for them. --- # The Inaugural Retail AI Index: Celebrating the Top 100 Brands and Retailers https://www.lily.ai/blog/retail-ai-index-100-brands-retailers 2024-07-31 The inaugural 2024 Retail AI Index, completed with Radii Group, ranks the top 100 brands and retailers and shows how AI is visibly reshaping apparel, footwear, beauty, and home. [image: The Inaugural Retail AI Index: Celebrating the Top 100 Brands and Retailers] The inaugural 2024 Retail AI Index is here! This comprehensive analysis, which was completed in partnership with Radii Group, shows how AI is visibly reshaping retail. By measuring the public demonstration of AI-powered innovation across leaders in apparel, footwear, beauty, and home segments, this index serves as a critical benchmark for understanding the integration and consumer-facing impact of AI in retail. [image: Excerpts from The Inaugural Retail AI Index] Within its pages, you’ll find the definitive ranking of the top 100 brands and retailers who are investing in AI and implementing it to create better experiences for their customers. All 100 of these businesses have very unique—yet extremely effective—implementations of AI technologies, as these innovations rapidly evolve both the world and the retail industry, specifically. This comes at a time when the global market size of AI in retail is expected to grow from $7.3 billion to $29.45 billion by 2028. Yet, did you know that American retailers and brands are slow to adopt AI, in comparison to their peers across the globe? China leads in AI adoption, with 58% of companies having deployed AI, while only a mere 25% of American companies have. This Index explores the most innovative and impactful implementations of AI in the retail space across the U.S., and celebrates today’s leaders. As eMarketer states in their coverage of the Index, “The wide array of companies featured in Lily AI’s index shows that AI is increasingly a must-have tool in retailers’ arsenals, especially as they look for ways to deliver more personalized experiences, drive efficiencies within their organizations, and head off competitors.” With all that in mind, let’s break down the numbers you’ll explore in the report. ## Discover 100 Retail Trailblazers Embracing AI Through a comprehensive analysis of publicly-available data surrounding companies’ investments in AI, we discovered the top 100 innovators in the retail space across fashion, home, and beauty categories. While AI itself is not new and has been utilized in retail for decades, recent advancements are opening up exciting new applications that go beyond enhancing operational efficiency, to further accelerating revenue and sales through innovations in product discovery and development. ## Get Inspired by 12+ Case Studies and 5 Impactful AI Applications How are the top 100 applying AI today? The Index dives into all of this and more, like the following examples of retailers and the technologies they’ve applied. [image: Examples of technology profiled in The Inaugural Retail AI Index] #### 1) Augmented Reality (AR) We detail what AR is, how it’s historically been applied to retail, and the innovations it’s ushering into the industry today. Lowe’s is a particularly strong innovator in the AR space, and in the Index, we detail how their Lowe’s Vision app helps customers navigate stores and find products based on their interests and shopping histories. #### 2) Machine Learning (ML) ML is the most popular AI-based technology deployed by retailers and brands. And this means that there’s a lot to dig into! The Index details some of its applications, with a special focus on how ASOS bakes ML into their shopping experiences, with its customer profile builder. #### 3) Deep Learning Amazon is famously innovative with their use of AI. The Index explores how they’re using complex deep learning systems to find patterns in customer behaviors and shopping preferences, reduce packaging waste, and find process efficiencies. #### 4) Real-Time Analytics One of AI’s superpowers is its ultra-fast processing time! Sephora is a retailer who is taking full advantage of this within the beauty industry. We cover how they’re using real-time data provided by online shoppers, in order to recommend the most relevant, interesting products. #### 5) Chatbots and Virtual Assistants As natural language processing improves, chatbots are also improving all the time! We dive into consumer preferences surrounding these helpful tools, specifically exploring how Williams-Sonoma has deployed a chatbot that’s designed to determine the perfect gift. ## Get All This AI Info and More! The worlds of AI-powered innovation and retail will continue to intersect. The top 100 honored within the 2024 Retail AI Index are at the forefront of this movement. At Lily AI, our vision is to bring humanity to shopping. Every day, we meet with retail industry leaders who share their strategies, challenges, and successes in delivering the best possible customer experiences. Inspired by these innovators and early adopters, and with our own disciplines as a data-driven and analytical company culture, we set out to quantify the state of retail AI today. Our Index and its findings place an emphasis on AI that goes beyond automation and efficiency, to measurably, meaningfully, and directly improve shopping ways that we each can personally experience as consumers. --- # When Is the Right Time to Implement Product Content Optimization for Descriptions, Attributes, and Metadata? https://www.lily.ai/blog/when-is-the-right-time-to-invest-in-product-content-optimization 2024-05-28 It is always the right time to improve your product data. But re-platforming, piloting, or building in-house can make it feel otherwise, so here is a plan for every stage. [image: When Is the Right Time to Transform Your Product Attributes?] To put it frankly, it’s always the right time to improve your product data and optimize product content! While this is a universal truth, each retailer has a unique position and situation that impacts how easy or difficult this task is. Retail tech stacks vary and are rarely static. You may be re-platforming, piloting, building a new solution in-house, or even eliminating aspects of your tech stack entirely. Variables like these may lead you to think that now is not the time to invest in Product Content Optimization (PCO), or frankly, you may not be sure where to even start! ## Poor Quality Product Data Drives Poor Performance Poor quality data equates to missed revenue opportunities. This is because AI-powered product content optimization enhances product discoverability, improves customer experience, and grows sales. In fact, AI can improve productivity in ranges from 5:1 to 100:1 or more, creating a $1.7 trillion opportunity for retailers. Additionally, the revenue opportunities presented by retail AI are undeniable, as 69% of retailers who use it report an increase in annual revenue, directly attributed to their AI adoption. Here’s your turnaround plan to supercharge advertising, marketing, and e-commerce by harnessing AI to optimize product content for every phase and stage. ## Product Content Transformation Guide #### Pre-Revenue or Pre-Launch? What wonderful timing! You can get started with Lily AI quickly and easily, building clean data and better customer experiences from the ground up. To get started, you’ll need to share product images and basic text that describes them. Lily AI uses a mix of generative AI, machine learning, deep learning, computer vision, and small language models to generate product content such as attributes and metadata from your text and images. It can also generate content based on CAD files or sketches, though photos and text generally provide more detailed outputs. The data and content Lily creates are highly detailed and inclusive of both objective product features as well as the natural language real people use when they search, shop, and speak. #### Re-Platforming? This is a critical time to kill two birds with one stone. You don’t just have to migrate your existing data to a new e-commerce, search, PIM or MDM platform. Instead, you can migrate enriched data into a new system and a new taxonomy. It’s absolutely critical that retailers have everything ecommerce-focused in order, as we see the worlds of AI-powered search and the ways people shop are changing. Upgrading your product data and taxonomy, while simultaneously replatforming, means that you’ll start fresh with massively improved customer experiences, product discoverability, and the ability to better connect with potential customers everywhere they search and shop. #### Don’t Have a Centralized Product Data Management Platform? Do you rely on an e-commerce platform alone, like Shopify? Retailers who do are in the perfect position to bolster their product attributes by way of AI. Whether you’re integrated directly into Shopify or a simple batch file via SFTP, we can help. Lily AI will provide you with clean, accurate, complete, and up-to-date data to import into any ecommerce platform or tool you use. For example, with our Shopify integration, brands and retailers can automatically export their product catalogs to Lily AI. The integration delivers attributes downstream, while simultaneously ingesting data directly from Shopify, without working with flat files or batch files. As a part of this process, retailers receive product attributes that describe physical product details, as well as trends, occasions, and styles. This makes it faster and easier than ever to enrich your entire product catalog with high-converting, customer-centric product attributes. The best part, Lily is tech-agnostic and can integrate with any system you use! #### Already Have Basic Product Data and Attributes? Let’s say you already have some basic product data and attribution capabilities in place . . . or you rely on vendor-contributed attributes throughout your store. In all likelihood, these attributes reflect the old adage “you get what you pay for”. How accurate and precise are they, and how consistent are the results? Do they capture basic details, or do they go deeper to capture more nuanced natural consumer language and preferences? Do you have the ability to define and refine the parameters that describe a particular aesthetic, or are you left with “out-of-the-box” definitions that don’t allow for customization? ## Lily AI: The Leader in Product Content Optimization Only a retail-focused AI that injects customer language throughout your product data and attributes will level up your product discoverability. Lily AI is the leader due to its robust, inclusive, and industry-leading dynamic taxonomy, which includes: - Objective Data: Including item types, materials - Subjective Data: Including styles, occasions - Synonyms: Such as “glossy” vs. “high shine” - Macro and Micro Trends: Such as Barbiecore, Modern Farmhouse, or Clean Girl - Customer Language Enrichments: Like site search queries, SEO terms - International Language: Including localized terms, translations The pace of AI can’t be beat either, as it consistently keeps product attributes, titles, and descriptions clean, accurate, and updated to reflect current customer language. Learn more about our industry-leading approach to Product Content Optimization and how it can impact the future of AI-powered search and shopping. --- # Report: How AI Bridges the Retail Consumer Expectation Gap https://www.lily.ai/blog/report-how-ai-meets-retail-consumer 2024-03-25 Our report with RETHINK Retail surveyed over 1,000 North American consumers across fashion, home, and beauty to address the gaps between consumer expectations and what retailers deliver. [image: Report: How AI Bridges the Retail Consumer Expectation Gap] “To say the quiet part out loud, there’s a widening disconnect in the world of retail today.” Our recent publication How New AI Strategies Are Helping Bridge the Consumer Expectation Gap doesn’t mince words in its opening line. This first statement, much like the entire report, directly addresses the gaps between consumer expectations and the reality of what retailers deliver. In partnership with RETHINK Retail, we surveyed over 1,000 North American consumers across fashion, home, and beauty to learn about these expectations, needs, desires, and interests. Inside the report, you will find details and data on what today’s consumers want from retailers and how those brands can meet their ever-rising, frequently evolving expectations. People are discovering products in new ways. They’re connecting with brands through a combination of social media, search, word of mouth, and more. With so much competition in the fashion, home, and beauty spaces—coupled with consumers feeling the economic crunch of inflation—it’s more critical than ever for retailers to understand who their potential customers are, how to reach them, and what to offer them. In this day and age, AI is the best way to meet these targets and move forward. ## Data-Rich Details on Shifting Consumer Expectations Our exclusive report delves into the heart of one of retail’s biggest challenges: using valid data to measure customer experience performance. Its focus is on actual consumer sentiments, rather than retailer assumptions. This makes our report a comprehensive analysis that reveals stark differences in how each group views today’s shopping journeys. By downloading the report, you’ll get insights into: - How people find the products they want: Are they searching on Google? Are they finding trends on TikTok? We explore the breakdown of how people find the products they actually purchase. - Details on where disconnects lie: In a world where 70% of customers quit their searches without buying, we uncover the why behind cart abandonment. We poll people on their personal frustrations with product assortments, search results, and quality upon delivery. - Personalization preferences: Our report examines how people prefer to receive product recommendations, interact with online shops, get discount offers, and more. - What consumers think and feel about AI: It’s hotly contested and consumers have mixed feelings about the technology . . . but AI can drive 90% revenue increases for retailers who embrace it. Here, we dig into these sentiments and learn how consumers really want AI to be applied to their shopping experiences. ## How Does Lily AI Bridge the Retail Consumer Gap? With all this valuable data on the modern consumer, one solution stands apart from others, when it comes to addressing their needs: artificial intelligence. Still, not every AI tool connects consumers with products, let alone drives revenue for retailers. Only vertical-specific AI with models that are trained on retailer data can bridge these gaps. As an AI solution built for retailers and trained on highly-specific industry data, Lily AI injects customer-centric language across entire retail ecosystems, delivering upwards of 9-figure revenue lifts for customers. Lily uses a combination of generative AI, computer vision, natural language processing, machine learning, and deep learning to elevate and enhance shopping experiences for consumers . . . and drive profitability for retailers. Lily AI addresses the need to optimize products for multiple search types that attract people to your digital storefront. As the report’s survey demonstrates, customers are finding products through both on- and off-site means, and Lily AI delivers enhanced product discoverability across other digital mediums, such as sponsored ads or product listing on Google. By bringing humanity to shopping through customer-centric language, Lily AI helps retailers understand who their shoppers are, what their priorities are, and how to serve them, ultimately bridging the consumer expectation gap. If you’d like to learn even more about how AI is ushering in a new era of retail, attend our webinar with RETHINK Retail on April 30. --- # C-Suite Questions About AI Answered https://www.lily.ai/blog/c-suite-questions-about-ai-answered 2024-01-29 55% of retail executives say they’ll invest in AI for marketing and 39% to enhance customer experiences. Answers to the cost, technology, and results questions the c-suite asks most. [image: Answering executives' questions about AI] - AI will have profound impacts on retail in 2024, as 55% of executives say they plan to use it to assist with marketing and 39% will use it to enhance customer experiences. - Executives who are concerned most about AI cost, technology integration, and results will find their most frequently asked questions answered below. - Learn more about building vs. buying AI, how to get the most out of your investments, and how they will augment your teams’ workflows to deliver stronger results. AI’s impact in 2024 will be huge for the retail industry, as it is the second-largest spender on the technology. This is because retailers who implement AI see a 14% boost. Many c-suite retail professionals have stated their intentions to invest in AI over time. In fact: - 55% of retail execs say they’ll invest in AI to assist with marketing. - 39% say they’ll invest in AI to enhance customer experiences. There are no signs that these investments will slow down, as AI drives revenue, enhances brands’ online presence, finds efficiencies, and fosters customer loyalty. Let’s look at how by answering retail executives’ questions about AI. ## A Cost-Focused Question About AI #### Will I need to hire new people when implementing AI? If you’re building an AI solution in-house, it’s most likely that you’ll have to hire new people. While an in-house engineering team may be able to complete portions of your build, you’ll likely need more specialized engineers, data scientists, a QA team, and a team for ongoing maintenance. If you purchase the right AI solution to augment your workflows, rather than building it yourself, your teams can continue their current projects with no interruption. ## Technology-Focused Questions About AI #### Is your data set large enough? If you’re building your own AI solution, you need to carefully consider whether you have enough data to work with, where it comes from, and if it’s truly specific enough to address your needs (which leads us to the next question). #### Is your data set specific enough? There are many all-in-one solutions out there that retailers need to be cautious of considering. While some of these solutions may promise quick results, their outputs may not be specific enough to meet your brand’s needs, due to their use of more general data, while excluding industry-specific and first party data. Moreover, these solutions may be inoperable with the rest of your tech stack. Ideally, retailers should work with AI solutions that seamlessly integrate with the other resources their teams already use. These types of AI products are available and should be a top priority when shopping for AI. In the long term, they offer more control and flexibility. ## Results-Focused Questions About AI #### How will AI impact my bottom line? Some of the biggest brands in retail—J.Crew, Bloomingdales, and thredUP—use Lily AI to drive greater sales and conversions and enhance personalized customer experiences. Beyond initial sale points, AI fosters customer loyalty, resulting in repeat purchases. #### How will AI impact internal processes? Processes that have historically been manual will be sped up. AI tags products earlier in the buying process and more accurately plans assortments, decreasing the amount of time it takes for new products to go live on your site. AI-assisted product attribution data has significantly more depth too, with up-to-the-moment trend language and a multitude of synonyms from Lily’s 25k+ product attribute taxonomy. Rich product descriptions are also instantly generated, saving your team time writing, enhancing description accuracy, and using the language of your customers. #### How will AI impact customer experience? Shoppers will be presented with items that truly match their interests, preferences, and personal style. As a result, on-site conversions and order sizes are increased and customer loyalty is fostered through enhanced site search, product recommendations, and product copy. ## Get Answers for Your Questions About AI Whether you’re ready to implement your own retail AI, or are curious to learn more about which solution is best for your brand, experts at Lily AI can help. Schedule a demo to learn how AI helps executives achieve the metrics they care about most. --- # How to Fast-Track AI Evaluation & Onboarding with Your Business Technology Team https://www.lily.ai/blog/business-technology-ai-eval-onboarding 2023-12-18 When implementing AI, effective collaboration with your Business Technology team is critical. How to convey business impact, ease of set-up, and high safety standards from the onset. [image: Business Technology Team Evaluating AI] - When implementing AI solutions, effective collaboration with your BT team is critical. - Share the criteria used to select an AI solution and show how it integrates with current technology investments. - Anticipate BT requirements, such as privacy and security as well as vendor onboarding resources. There’s huge promise—and huge hype—surrounding AI. When it comes to choosing their optimal AI solution, retailers need to verify that their choices meet the expectations of the CIO and BT team. After all, BT professionals ensure that their technology implementations drive the business forward, are worthwhile, totally secure, and safe to use. If you’re championing Lily AI for adoption within your retail operations, BT will complete its implementation. To get their collaboration started on the right track, we recommend you convey Lily’s business impact, ease of set-up, and high safety standards. Engaging with your brand’s BT team from the onset ensures that Lily AI is seen not just as a new piece of technology, but as critical to pushing your organization forward. ## Who is BT? Understanding Their Goals & Challenges Oftentimes, BT teams are unsung heroes of retail companies. They are critical drivers of business success, working behind-the-scenes to navigate our ever-evolving digital landscape. Cybersecurity, cloud computing, software development, and software integration generally fall under their umbrellas. BT keeps many business functions running, while ensuring the security and usefulness of all digital systems. Oftentimes, these teams are stretched thin, with such large responsibilities and fast-paced projects on their plates. It’s crucial to recognize that BT teams aren’t uniform. Their roles vary from organization to organization. Considering this, take time to learn who your own internal BT operations consist of and what their responsibilities are. This will help you know which aspects of Lily AI to discuss with them. When implementing a solution like Lily AI to your BT team, meet them with empathy. Show that you understand their workloads and responsibilities. Discuss how Lily AI is low-lift for them, while still delivering a substantial revenue impact. ## Lily AI Makes the Most of Your Current Technology Implementing Lily AI is easy because it augments the software retailers already use. It makes your current technology more efficient and effective, maximizing existing tech stacks to deliver 9-figure revenue lifts. When communicating the benefits of Lily AI to your BT team, explain how Lily injects customer-centric language en masse, across your entire PIM, improving the following functions: - Site Search - Google Search - Product Recommendations - Item Setup - Demand Forecasting With Lily AI, there’s no need for retailers to build their own AI solutions. Augmenting your retail technology suite with vertical AI offers a tailored solution for your brand that is customized and faster to implement than building anything. Pro Tip: Lily AI augments your tech stack, making your current technology more efficient and effective, resulting in 9-figure revenue lifts. ## Implementing Lily AI is Fast & Easy Disrupting services and processes when integrating new technology is a common hesitation of BT teams. Lily AI addresses these concerns with its simple implementation process. In fact, there’s no need to change much at all. Explain to your BT team that Lily eliminates the need to “rip and replace” any tech that your teams currently rely on. Its implementation is non-disruptive to ongoing workflows. Retail BT teams will implement Lily AI’s pixel on your website. This gathers granular data and insights from every stage of your shopping funnel, analyzing customer habits and journeys, while not impacting load times or website experiences. It facilitates real-time updates for e-commerce performance tracking, while requiring minimal ongoing support and reducing BT resources needed for ETL and exporting. The implementation process is designed to be as simple as possible for BT teams. Lily AI staff members will work with a dedicated Tech Operations professional at your company during onboarding. This process offers collaborative learning and troubleshooting experiences, so your users and BT experts are set up for success. Pro Tip: Lily AI’s simple implementation process means that there’s no need for your BT team to “rip and replace” any software your company is currently using. Implementation is easy and an internal BT teammate will have a seamless, collaborative onboarding experience with Lily AI. ## Feel At-Ease With Lily AI’s Security Perhaps above all, security is a top concern for BT professionals, and Lily AI understands the importance of instilling confidence in its users. Lily AI is SOC2 compliant. This compliance meets the highest standards of security over a prolonged period of time, ensuring that all data is stored and processed safely. Additionally, Lily AI does not collect Personally Identifiable Information (PII). This minimizes security concerns for both retailers and their customers, which is an additional benefit to relay to your BT team. Pro Tip: Communicate Lily AI’s commitment to security and SOC2 compliance, which can ease any apprehensions of your BT team and foster a sense of trust and confidence. ## Implement a Results-Driven Retail AI With BT’s Assistance Engaging your Business Technology team from the onset of your journey with Lily AI means that your organization’s most tech-savvy professionals will experience easy onboarding and continued support. With their assurance, leadership will trust that you are championing a safe, secure, and effective AI solution to push your business forward. --- # How to Effectively Co-Champion Technology Investments with Your Data & Analytics Business Partners https://www.lily.ai/blog/win-data-analytics-experts-lily-ai 2023-11-28 Lily AI can significantly augment decision-making and operational efficiency, but championing it means conveying that potential to data and analytics professionals from the onset. [image: LIly AI's impact for data & analytics professionals] - According to PWC, 53% of CFOs say they plan to accelerate digital transformation using data analytics, AI, automation, and cloud solutions. - Bring your data & analytics team into Lily AI’s process early, testing a hypothesis as you pilot, in order to measure impact and see how Lily fits into your CAO’s agenda. - Dual-track both short-term and long-term initiatives to quickly gain traction and prove value while building the foundation for higher impact outcomes. In the evolving landscape of business technology, Lily AI stands out as a transformative tool that can significantly augment decision-making and operational efficiency. For those championing this innovative solution, it is crucial to effectively convey its potential to data and analytics professionals. Engaging with these experts from the onset ensures that Lily AI is seen not just as a new piece of technology, but critical to organizational success. In a world where 53% of CFOs plan to accelerate digital transformation using data analytics and AI, getting your analytics teams’ buy-in on Lily AI is a pivotal asset in the quest for data-driven excellence and optimal business outcomes for retailers. ## Bridging Business Goals with Analytical Precision Analytics professionals often categorize business initiatives on two axes: - The x-axis represents long-term strategic goals with the potential for substantial business impact. - The y-axis focuses on short-term objectives with immediate benefits, which are relatively smaller in scale but easier to validate. It’s equally vital to position Lily AI within both of these short and long-term perspectives. In today’s era of big data, marrying human intelligence with machine learning is essential. To someone immersed in data and analytics, Lily AI might primarily appear as a data service. The most straightforward approach to gain a data scientist’s support is to demonstrate the return on investments (ROI) of this data, ensuring its inclusion in ongoing budget plans. However, the conversation shouldn’t end there. Let’s dive into all of Lily AI’s other proof points to prove its value to data and analytics professionals. #### Immediate Benefits & Strategic Longevity Lily AI is uniquely positioned to address both the x-axis of long-term, strategic imperatives and the y-axis of immediate, short-term advantages. Consider the rapid wins achievable through digital use case enhancements, like site search optimization. This not only delivers quick benefits, but also sets the stage for wider e-commerce applications, such as assortment planning and demand forecasting. Lily AI excels in translating such targeted technologies into substantial, measurable business gains. #### The ROI Imperative For data scientists and business analysts, the value of a new tool lies in its ROI. Demonstrating Lily AI’s ROI is straightforward. Its advanced analytics capabilities translate into direct budgetary benefits. In fact, you can read several Lily AI case studies that demonstrate value that was directly added after adopting this technology. Yet, the strategic narrative should extend beyond immediate ROI. It’s about embedding Lily AI in the long-term growth strategy of the organization, recognizing that it complements human analysis with machine precision—vital in an era where retail is inundated with data. ## Enhancing Collaboration With Data & Analytics Professionals #### Address the Redefined the Role of CAOs The role of the chief analytics officer (CAO) has evolved from being a technical position to a more strategic leadership capacity. The modern CAO is a master of turning data into valuable business decisions. The role demands making astute business and investment choices, constantly aiming to optimize operating costs, and doing so judiciously while fortifying the organizational structure. When seeking buy-in from CAO’s, understand that these individuals are looking at and interpreting a spectrum of data, including: - Descriptive Analytics – What happened? - Diagnostic Analytics – Why did it happen? - Predictive Analytics – What might happen? - Prescriptive Analytics – What should we do? By understanding these stages, you not only immerse yourself in the analytics mindset, but can tailor your pitch to resonate with these data-focused executives. In promoting Lily AI, it’s important to resonate with the CAO’s expanded remit, emphasizing how the platform supports a range of analytical tasks—from descriptive to prescriptive. Pro Tip: Use the language of analytics to expedite understanding and alignment. #### Crafting a Cohesive Data Narrative When advocating for Lily AI, it’s essential to understand its role in the broader analytics landscape. A business analyst, who often has a broader business perspective, can be your strongest ally here. Navigating the transition from conceptual thinking to tangible results requires more than just a vision—it demands strategy, synergy, and a thorough understanding of the analytics landscape. Illustrate how Lily AI offers not only immediate performance improvements but also aids in scaling analytics for long-term strategic benefit. Major initiatives succeed when phased thoughtfully, balancing quick wins with sustained, strategic impact. It’s a balancing act between immediate achievements and long-term goals. Identify and speak to the immediate, tangible wins while concurrently crafting a broader, impactful long-term plan. While ROI is a crucial metric, it’s not always black and white. In technology investments, especially AI, the focus should shift from only ROI to broader, sustainable value. When speaking to data analysts, emphasize AI’s potential to deliver enduring, incremental benefits, validated through methodical analytical techniques that can be tested. Pro Tip: Engage your business analyst early on. Collaborate to build a robust business case for data scientists. #### Drive Insights with a Hypothesis-Led Approach Lily AI thrives on its ability to merge raw data with human insights. Starting with a well-defined hypothesis, it facilitates a collaborative exploration of insights that inform cost-benefit analyses and drive business actions. This marriage of data, situational context, and domain expertise is where profound insights can emerge. It’s critical to begin conversations with the data team to convey the importance of starting with a distinct objective and a hypothesis that can be confirmed or debunked. Pro Tip: Encourage a culture of data transparency to fuel a diversity of insights and foster innovation. ## Charting a Data-Driven Future with Lily AI Innovation can often be stifled when organizations aim to internally replicate technologies. Lily AI represents not just an addition to your technological suite, but a shift towards an integrated, data-fluent organization. Engaging your analytics team from the onset not only aids in pinpointing the right problems. This also crafts solutions that deliver on both immediate and strategic goals. Lily AI is about understanding the full picture—how data can be leveraged for quick wins and how it can underpin more significant transformations. By approaching your data and analytics team with a blend of technological comprehension and strategic foresight, you pave the path to a future where data isn’t just a resource—it’s the backbone of business innovation. --- # Trend Attribution 101: Capitalizing on Cycles Sooner https://www.lily.ai/blog/trend-attribution-101-capitalizing-on-cycles-sooner 2023-05-17 It’s not enough to identify a trend. A look at the fads, mainstream, and classic trend cycles, and how AI-powered trend attribution makes current trends actionable today. [image: Trend Attribution 101: Capitalizing on Cycles Sooner] Key Takeaways: - It’s not enough to identify a trend–it requires AI-powered trend attribution of your product catalog to capitalize on consumer demand earlier in the trend cycle. - As opposed to trend prediction and demand forecasting for next year or next season, trend attribution makes current trends actionable today. - With Lily AI’s product attribution, retailers and brands can instantly curate collections and launch promotions faster to capitalize on trend cycles sooner. At Lily AI, we know that different consumers shop differently. Whether casually browsing or ready to buy, they’re shopping with a preference or need in mind. This can often be by occasion, activity, style, fit, feature, or trend. Of these categories, trends can be the most difficult for retailers and brands to act on quickly. That’s not because retailers and brands aren’t in the know about the latest trends. It’s because some trends, like “fads”, surge and peak so quickly that it’s difficult operationally to keep up. Sometimes, products are attributed with relevant trends during item setup, but the trend data isn’t continually refreshed and becomes out of date. The worst case scenario is that trend attributes are ignored or missed entirely. Let’s explore the different trend cycles, and see how AI-powered product attribution can help retailers and brands turn trends into more revenue. ## Understanding Fashion, Home and Beauty Trend Cycles When a customer purchases and wears a certain style or buys a product based on a trend, that style is then considered “accepted.” However, social media has changed the trends landscape to the point where many fashion and beauty trends DO NOT always survive and have a literal shelf-life. Lily AI, the product attributes platform, identifies three types of trends: Fads, Mainstream, and Classic. Fads: - Essentially “a craze” – i.e., trends that only last for 1 month or season. - Typically inspired by social media influencers, celebrities, and even designers. - Trends that may not appeal to the average consumer. - May include novelty items or unconventional, statement pieces. Mainstream: - “Ordinary” clothes that most consumers may wear for multiple seasons. - Nothing too outlandish or extreme. - Conventional, trendy, and relevant. Classic: - Classic trends last for several seasons – and sometimes years – to come. - Timeless, ceases to go out of style. - Sustainability and quality are key indicators. [image: The fashion cycle] How does the home cycle differ from the fashion cycle? In general, home items are costly to produce in comparison to fashion or beauty items. Subsequently, the home trend cycle moves much slower. Decor and soft furnishings (pillows, sheets, towels, etc.) are the first items to be adopted. Unlike hard furnishings, they can be easily swapped out. ## Capitalizing on Trends Faster How do you curate a trend-based collection or campaign on the fly? By enriching the entire product catalog with more robust attribution data that includes trends. With these trend attributes ready to go, e-commerce and merchandising teams can quickly and easily build trend-based thematic collections and landing pages. Marketers can then take these collections a step further by integrating them into campaigns and promotions. For the hottest trends, Lily AI automatically provides trend attributes. Need something more custom? Our team of domain experts can work with our clients to define and apply a custom style or trend tag to their catalog. By creating collections and campaigns faster, retailers and brands can capitalize on trends earlier, maximizing the revenue potential for every trend. The best part is that your customers will love being able to easily shop the latest trends, especially Gen Z and millennial shoppers that follow social-media inspired microtrends. Learn more about the many benefits of AI-powered product attribution and how it can transform your business, elevate the customer experience, and maximize revenue. --- # Thrift the Look with Lily AI’s Fashion Product Attribution https://www.lily.ai/blog/thrift-the-look-with-lily-ais-fashion-product-attribution 2022-08-17 Online thrift shopping keeps growing, so for National Thrift Shop Day we’re sharing our top tips for making the most of resale in retail. [image: Learn how thredUP and Lily AI help shoppers “Thrift the Look"] The concept of thrifting clothes is not a new one and it’s more than likely you can find numerous thrift stores in your own neighborhood that are just waiting to be explored. However, along with physical thrift stores, the presence of online thrift shopping has been growing and with National Thrift Shop Day amongst us, we wanted to share our top tips for making the most out of resale in retail and thrift shopping online. ## Tip 1: Know Your Personal Style One of the best things you can do before choosing to shop online is to identify and understand what your own personal style is. Knowing your signature style can help to serve as a filter, allowing you to shop more strategically and intentionally. When you buy pieces that you know you’re going to absolutely love that are not only going to flatter your personal style, but also represent your personality, that’s when you know you’ve captured the vision of your best self. ## Tip 2: Select Your Color Palette and Know What You’re After Shopping online can certainly feel overwhelming at times so nailing down concepts like your own unique color palette and which key pieces of clothing are missing from your wardrobe can help to prevent you from rushing to fill a gap or getting sidetracked and purchasing something you don’t really need. A key benefit of knowing your preferred color palette is that this can help you stick to the colors you know work best for you, permitting you to maximize wear and establishing more versatility with the ability to mix and match items much more easily. ## Tip 3: Choose a Shop with Spot-on Site Search Results Even if you know what you’re looking for when you go thrift shopping online, it’s going to be next to impossible if the search results aren’t bringing you the pieces you actually want. In order for you to be able to find what you’re actually looking forwith the least amount of clicks possible, then 10+ pages of irrelevant results isn’t going to cut it. When an online shop uses granular, customer-centeredproduct attribution data and enables long-tail contextual searches, you can easily be guided directly to the items you really want without the hassle of endless scrolling. ## Tip 4: Leverage a Tool Like thredUP’s “Thrift the Look” Using an online tool with shopping can help you to remove some of the noise and prioritize what works best for you by finding the right matches to your sizing, price range, brand preferences, and even color palette. thredUP, the world’s largest online secondhand shopping destination with thousands upon thousands of like-new styles from favorite brands around the world at up to 90% off the price of estimated retail, functions similarly to most online shopping platforms. However, they put themselves high above the rest with their tool, “Thrift the Look.” This innovative tool, co-developed by Lily AI allows you to recreate the outfits you love sustainability, matching your most-hearted looks with similar styles. With more than 100,000 items arriving every day from sellers’ closets, thredUP works to categorize and tag products at tremendous scale. Its inventory includes more than 50,000 brands and hundreds of categories, each with dozens of attribute groups—any of which could lead you to the item you want. Lily AI’s Product Intelligence helps thredUP shoppers like you to find what you’re looking for more exactly with two to three times as many attributes to search and explore for the perfect outfit. On average, Lily AI has increased sell-through by 15%, along with - Tagging 275,000+ images per week - Providing a 2%+ conversion rate lift for customers who’ve purchased >2 times - Producing a 2.8%+ conversion rate lift on iOS - Increasing favorites per user by 2.1% - Co-developing the cutting-edge “Thrift The Look” shopping experience [image: Purva Gupta on thredUP] --- # The Top 3 Tips for Transforming Product Discovery https://www.lily.ai/blog/the-top-3-tips-for-transforming-product-discovery 2022-07-18 Even with all the right products, you have a very real problem if shoppers can’t locate them. Our top three tips for transforming online product discovery. [image: Man smiling at his phone while discovering new products while online shopping.] As an online retailer, it’s a matter of fact that even if you have all the right products, you have a very real problem that needs addressing if shoppers can’t actually locate those products. This is where better online product discovery makes a big difference. To help move the product discovery process forward, here are our top tips: ## Tip 1: Keep Product Data Detailed with the Power of AI To keep product details healthy and robust, rather than thin and inconsistent, it can be an effective method to deploy AI-powered image recognition in providing an assist. People each have their own ways of searching for items, and that doesn’t always fit with the cut and dried attributes that may have originally been associated with a product by a merchant or a manufacturer. Humans are more intentional and if they’re looking for a New Year’s Eve dress, then that’s exactly what they’re going to search for. However, they’re not going to have much luck finding their dream dress if that item wasn’t set up and attributed that way. With the help of AI-powered image recognition, more product attributes can be extracted and the chances of that more exact match (i.e. NYE dress) being found go up. This allows retailers to boost product discovery and conversions. Brands and retailers who merely take in the generic product attributes provided to them by manufacturers and distributors – and don’t also implement AI that takes the language of the customer into consideration – aren’t going to successfully allow shoppers to discover the products they’re looking for. [image: A woman in a black sequin NYE dress] ## Tip 2: Make Items Quick and Easy to Find In order for a shopper to be able to find what they’re looking forwith the least amount of clicks possible, it isn’t going to be quick or easy if the results returned are 10+ pages of irrelevant results. To keep results relevant for shoppers, the results need to match their shopping goals. By using granular, customer-centered product attribution data and enabling long-tail contextual searches, retailers can easily guide shoppers directly to the items they actually want without the hassle of endless scrolling. ## Tip 3: Implement Better Personalization Practices Quick and easy product discovery not only brings results that are more likely to be relevant, it also can help with better personalized recommendations. In order to show the shopper that you know and understand who they are as an individual, this requires an understanding of not just what they purchased – but why. When your taxonomy is rich with details from AI, it learns about the customer and can help retailers understand how they engage on a website. This helps to create product recommendations that reflect a customer’s genuine personal preferences, leading to higher conversions. --- # 6 Steps to E-Commerce Site Search That Converts https://www.lily.ai/blog/6-steps-to-ecommerce-site-search-that-converts 2022-06-09 When shoppers type a specific term into a retailer’s search bar, they too often meet irrelevant or blank results. This guide covers the challenges and six steps to better on-site search. [image: 6 Steps to E-Commerce Site Search That Converts] When consumers head to their favorite retail e-commerce website and type a typical, specific search term for something they need or want into the site’s search bar, what do they typically find? If you’re a shopper yourself – and you almost certainly are – you know that it’s very often a very suboptimal experience. Shoppers, even on the websites of the world’s leading brands and retailers, are met with a surprising amount of resistance, in the form of multiple pages of irrelevant results; search results that are close but no cigar; and sometimes, completely blank results pages. It can be a fun and sobering exercise to try it out yourself. A fun exercise if you’ve got some time is to conduct a search like this of your own on a few retail e-commerce sites. We made one of our Lily AI co-workers conduct a search for the type of shirt he usually wears to work, a search that would probably go something like this: “collared untucked checked men’s large.” At one high-profile retailer that very much sells men’s clothing in this style, there appeared to be no collared, checked men’s shirts – even though he knew they sold them – yet he did receive these “very helpful” recommendations for items that he might like instead: What do you think the chances were, had this been real, that his time on this site would have ended right then and there? The point is that every consumer has their own “long-tail” way to describe what it is they’re looking for. As we make our way in the world, we learn different colloquialisms, expressions and features that we use as shorthand to describe what’s important to us. Yet this “search diversity” is barely acknowledged on a high percentage of modern e-commerce retail sites, and the upshot of this problem is missing results from on-site search that, had they been returned to the curious shopper, could have led to an instant purchase. There’s also the shopper frustration that comes from being recommended hand towels and teacups when the goal is to find a new shirt to wear this Monday. This guide is for those retailers and brands who understand that there are huge gains to be made in both revenues and customer satisfaction in getting this right. We’ve outlined some of the current challenges, as well as 6 potential solutions that will help the modern e-commerce retailer make on-site search one of the very best things they do, as opposed to a liability that drives customers away. Read on! > 68% of shoppers say they won’t return to a site that provides a poor search experience. ## Search is Where Retail E-Commerce Begins We now know that search has become absolutely central to the retail e-commerce experience. Up to 43% of visitors to retail websites immediately go to the search bar, and those consumers are 2-3x more likely to make a purchase than site visitors who don’t use the search bar. (1) In fact, we have a few other stats that underscore the importance of search: - Search bar shoppers spend 2.6x more money than shoppers who don’t use the search bar. (2) - 68% of shoppers say they won’t return to a site that provides a poor search experience. (3) - While the average conversion rate across all websites measured at 2.77%, site search users converted at 4.63%. (4) And yet over 60% of sites require users to guess the exact “correct” search term for a product. (5) Let’s take a look at why this is so frustrating for end users, and why it runs so counter to the notion of search diversity. ## Different Shoppers Shop Uniquely Consider the synonyms for “Loose Dress” in this example. What you might call a loose dress I might call a sundress; your best friend might call it a nap dress and her mother might call it a house dress: The key to serving all four of these individuals is to build a robust product taxonomy that takes into account how real people search in the real world. Sure, those who are already focused on a particular brand name can often unknowingly filter their search just to the results from that particular brand (i.e. Vera Wang, Hugo Boss, Topman etc.), and that certainly makes it easier. For the majority of shoppers looking to compare items between brands, however, it’s essential to assign attributes to items that allow for the rich and varied diversity in what we actually call things, and what’s truly important to us when we shop. Any initiative to improve the digital experience must keep search front and center. Improvements to the search experience can make the most impact because they directly affect what we call high-intent shoppers – the ones who know what they want, but who may describe what it is they’re looking for in a “language” different from that used by the merchant they’re visiting. If these retailers can instead serve up relevant and comprehensive results for their most promising site visitors, they’ll convert more orders and build brand loyalists. The stakes couldn’t be higher. PwC reports that 32% of customers will leave a brand they love after just one bad experience. (6) Now lets start talking about solutions – the 6 Steps to Better E-Commerce Search that you can begin implementing right away. ## 6 Steps to Better E-Commerce Site Search Nearly 84% of companies don’t actively optimize or even measure their on-site search, which points to a true opportunity in the marketplace. The retailer who does invest some time and money into optimizing site search can stand out quite quickly. Here are the 6 steps to undertake to make sure your organization is the one that does. ### 1. Understand the Current State of Your On-Site Search You can’t measure improvement until and unless you understand your starting point, so begin with a cold-eyed analysis of your site’s current search capabilities. Most e-commerce site search fails on a number of fundamentals. The biggest takeaway, as mentioned previously, is that over 60% of sites require users to guess the exact “correct” search term for a product. Most sites fail to recognize (and account for) typos, different phrasings, colloquialisms, contractions or symbols and synonyms. And all too often, on-site search fails to deliver hyper-specific results from specific, long-tail searches (even though there are a handful of products in a retailer’s catalog that would fit the criteria). There are several ways to assess how efficiently your search works and potential inadequacies. One is to look at your on-site metrics: If you’re seeing a high level of bounce after one or two searches, that’s a red flag. Another is to create a routine that flags searches that return few or no results, which gives a clear insight into specific searches that don’t work. A third and very important option is to pay actual shoppers — not your web team or your marketing team — to search the site for specific items, noting both how successful they were and how much effort it took. Most importantly, ask them what they liked and disliked about the process (and how it could be improved). ### 2. Offer More Impactful Filters and Facets Adding filters and facets to your search tool and product category taxonomy is one of the most powerful ways to enhance your product discovery capabilities. Filters allow your users to choose applicable options from a list you provide them: manufacturer, color, product category or anything else that might be meaningful — getting them closer to the products they’re looking for. Don’t overlook the fact that this also allows users to filter out specific characteristics. It’s more important to deliver accurate results than to get as many products in front of the customers as possible. A search that returns 10,000 hits is almost as useless as one that returns none, so this is important. ### 3. Add New Categories On a similar note, you should categorize your products as much as possible, and in as many ways as possible. For example, dresses could be categorized by style, color, size range, type of fabric, embellishments, or any other attributes meaningful to your product catalog. You should also think in terms of thematic searches, something only about half of the companies in Baymard’s research handled well. These categories are drawn conceptually rather than by the characteristics of the garment itself. Examples might include “spring outfit,” “party dress,” and “work shoes.” These categories, and the searches and toggles mentioned previously, work together to speed and empower search. For example, the phrase “work shoes” might mean very different things to a nurse, a factory worker or a senior partner at a law firm. Adding a well-thought-out set of filters to the basic “work shoes” category allows your users to further filter search results based on their style preferences. ### 4. Use Deep Tagging to Broaden Searchable Data Points One limitation you’ll inevitably encounter in your quest to improve product discoverability is inaccurate and incomplete product data. Sure, you can simply import the primary data points from your suppliers – what are often called “legacy, out of the box attributes” or general attributes, sourced from manufacturers and distributors. From that base level of searchable characteristics — size, style, color, fabric and so on — you decide how many additional tags or attributes are attached to the product. But this is often done manually, so it’s costly and prone to inconsistency and inaccuracy. The more products you have, the more quickly you’ll reach the point of diminishing returns. Lily AI’s product attributes platform breaks that limitation. Our proprietary machine-learning algorithm analyzes your product line using the language of customers (not merely suppliers, distributors or merchants) and applies deep tagging, potentially creating dozens of new data points for your products. Better yet, because it’s automated, this isn’t a massive, time-consuming project. Your site’s search capabilities will be able to utilize all of this new data to deliver accurate search results based on long-tail, subjective and objective searches within just a few weeks. ### 5. Make Better Search Autofill Suggestions At this stage of the process, with the high-level “broad strokes” fixes in place, you can begin to focus on finesse. One way to do this is through the use of improved autofill suggestions in your search tool. Your search tool may base its suggestions on site-wide search history, so a user typing in “spring dress” will see suggestions for all spring-related categories after the first word but only spring dresses after the letter “d” is typed. Autofill and autocomplete are much more powerful with an expanded, relevant product taxonomy (“the language of the customer”), because now you’re able to make suggestions that capture the many long-tail searches that your shoppers actually use, then deliver them exactly what they’re looking for: “spring dresses easter”, “spring dresses floral pattern” and so on. It’s a virtuous cycle. It makes users more confident in the search and encourages them to add more details, leading to more precise matches and better search conversion overall.” ### 6. Iterate, Iterate, Iterate An important point to remember is that improvement and optimization of your website is not a one-and-done task to be checked off of your list, but an ongoing process to build into your workflow. Each step in this process informs the others: Reducing bounce gives you more user searches and interactions to observe; user feedback helps you improve your search tools; clickable filters and categories tell you about your users’ search priorities; deep tagging empowers the creation of more finely tuned filters; and so on. As you repeat this process, your grasp of your users’ search process will improve immeasurably. You’ll know which filters they use, which search tools they find intuitive and which changes to your search made the biggest differences to your conversion rates and sales figures. It’s a virtuous cycle, and a very powerful one. ## The Bottom Line: Superior Product Discovery That Converts It’s no longer acceptable to return 10-plus pages of results that drop off dramatically in relevance after the first click. If you read customer ratings and monitor mentions of your company online and on social media (and you should, it’s important), you’ll know when your efforts are bearing fruit. How? Because your customers will rave about your brand. They’ll share, they comment, they show excitement. They’ll convert where you want them to – on your website, not someone else’s. It’s not just in search, either. Speaking in the language of customers, rather than that of manufacturers or distributors, has hugely beneficial ramifications across the entire retail value chain. A product with two to three attributes can, through visual, AI-driven tagging and a carefully-designed taxonomy, immediately unlock its sales potential to become a product with 10-15 attributes. When product attribution is done correctly in this manner, it instantly drives better site search, filters and facets, product recommendations, and demand forecasting. On-site conversion goes from an anemic retail industry average of 2.5% to 4%, to 5% and more – and just as importantly, the ability to sell at full margins is enhanced because demand is now forecast using that language of customer-driven attributes – not legacy, out-of-the-box attributes. This increases the ability to sell to customers what they’re looking for right now, and decreases the need to mark that inventory down later. “It’s the way we’ve always done it” becomes a thing of the past, and retail customers have a new set of forward-looking, customer-focused brands to pin their shopping allegiances to. ### Sources 1. Neil Patel, 2021, https://neilpatel.com/blog/site-search-killing-your-conversion/ 2. Neil Patel, 2021, https://neilpatel.com/blog/site-search-killing-your-conversion/ 3. Forrester Research, https://www.addsearch.com/blog/shockingly-high-cost-poor-site-search/ 4. eConsultancy, Is site search less important for niche retailers?, https://econsultancy.com/is-site-search-less-important-for-niche-retailers/ 5. Baymard Institute, Deconstructing E-Commerce Search: The 8 Most Common Query Types, https://baymard.com/blog/ecommerce-search-query-types 6. PWC, Experience is Everything: Here’s how to get it right, https://www.pwc.com/us/en/advisory-services/publications/consumer-intelligence-series/pwc-consumer-intelligence-series-customer-experience.pdf --- # The Top 6 Tips for Improving Site Search https://www.lily.ai/blog/the-top-6-tips-for-improving-site-search 2022-05-25 Search is where modern retail e-commerce begins. A look at the most common on-site search challenges, and a preview of our top six tips for overcoming them. [image: Man with glasses smiling at his phone while searching a retail site for products.] Search is where modern retail e-commerce begins, so getting it right holds the key to higher revenues, happier customers, and an experience that keeps shoppers coming back time and time again. When on-site search is done well, retailers and brands can enhance their overall site experience so that search works with them, not against them, to become one of the top ways they’re driving revenue growth. In order to optimize site search for shoppers, let’s take a look at some of the most common challenges e-commerce sites deal with and how to overcome them with a preview of our top six tips. ## Challenge #1: Search Diversity Different shoppers shop uniquely. For instance, while one shopper may refer to a dress as a “sundress,” another just might solely consider it a “loose dress.” Or, what one person may refer to as a “blazer” may be more well-known as a “sport coat” to someone else. There are so many attributes that can go along with just one item of clothing and keeping this level of synonymous search diversity in mind is key. ## Challenge #2: Irrelevant Results If a shopper is trying to find a unique item and irrelevant results pop up, frustration is going to arise. For example, if a shopper searches for a “sleeveless lace top” and instead gets results for a “lace hand towel,” those results are essentially going to be deemed irrelevant (and irritating) to the shopper (unless they’re really in need of new hand towels). ## Challenge #3: “Product Not Found” In a worst case scenario, a shopper looking for a specific product that doesn’t fit the terminology you have on hand, may receive the dreaded “Product Not Found” response. This is bad news for your e-commerce brand as shoppers that can’t find what they’re looking for are not likely to stick around and keep looking. They’re going to move on to the next site – a site with more optimized site search capabilities and better product attribution data built with the language of the customer at the top of the clothing pile. ## Tip #1: Know the Parameters of Your On-Site Search There’s no other way around it, you can’t measure improvement unless you have a clear starting point established. Start by analyzing your site’s search capabilities and assessing on-site metrics like bounce rate, flagging searches that return few or no results, and including actual shoppers when it comes to evaluating the effectiveness of a site’s on-site search. ## Tip #2: Focus on Impactful Filters and Facets One of the most powerful things you can do as a retailer is to enhance e-commerce customer intent with better product discovery capabilities and one of the best ways to do this is by adding filters and facets. Filters are broad categories (i.e., a specific designer/brand), while facets are a bit more specific (think, the color “gold,” or only size small). Including impactful filters and facets to your search tool and product taxonomy can be used to narrow down a set of items for shoppers so they can more easily find the specific thing they’re looking for. Or, they can be used to widen the search. Overall, they’re a great addition to on-site search and serve as strong ways to help users help themselves. ## Tip #3: Create New Categories It’s no surprise that a dress may be categorized by style, color, size, or any other relevant attribute that may serve as meaningful within your product catalog. But consider expanding even further by creating new categories. For example, in terms of thematic searches, creating categories such as “New Year’s Eve dress” or a “Cocktail party dress,” or maybe even a “70’s-style dress” for an upcoming 70’s-themed bachelorette party. Including a wide variety of well-thought-out thematic filters can make an immense difference. After all, a “bachelorette party dress” can look very different depending on the theme – e.g., boho chic, 20’s vintage, or Y2K. [image: ] ## Tip #4: Utilize Deep Tagging Legacy, out-of-the-box attributes sourced from manufacturers or distributors are highly limiting, costly, and prone to inconsistency and inaccuracy. Utilizing machine-learning algorithms to create “consumer-driven attributes” that use the language of the consumer are much more effective and can potentially create dozens of new data points for your products that would not exist in the supplier/distributor/manufacturer-driven world of simple language. ## Tip #5: Amplify Search Autofill Suggestions Don’t be afraid to get even more personal based on an individual’s own search history, amplifying autofill suggestions by taking their past search results into account. ## Tip #6: Iterate! It’s important to remember when tackling on-site search that working to improve it is an ongoing, consistent process. Each step is key in informing the next and know that when you work to repeat the trustworthy process, your grasp on each shopper’s on-site experience and search process can continue to be improved beyond measure. --- # The Step-By-Step Guide to Using Enhanced Product Attribution Data https://www.lily.ai/blog/the-step-by-step-guide-to-using-enhanced-product-attribution-data 2022-03-10 Retailers have underinvested in a critical area: their own product data. This guide covers seven ways to elevate the customer experience with AI-powered product attribution. [image: The Retailer’s Step-by-Step Guide to Using Enhanced Product Attribution Data] In the retailer’s quest to delight customers by investing in consumer data and insights, they have underinvested in a critical area: their own product data. By harnessing the language of the customer to better describe their products, retailers can transform merchandising, e-commerce, and marketing–and deliver more enjoyable shopping experiences that drive sales and build loyalty. In this Retailer’s Guide, we’ll show you the 7 ways to get started with AI-powered, customer-centric product attribution! ## Get a Preview of the Guide’s 7 Ways (Warning - Spoilers Ahead!) ### Supercharge Site Search The missing link in your CX and site search strategy isn’t a search platform, it’s customer-centric product attribution. ### Boost Personalization and Recommendations Nothing says “you get me” like relevant product recommendations, yet effective personalization is easier said than done! ### Optimize Filters and Facets Filters and facets are a core part of every website, and for good reason–they make shopping easier! ### Create Product Descriptions that Convert Improve PDPs by highlighting all of the most shopper-friendly and search-optimized product attributes. ### Amplify SEM and SEO Help shoppers find precisely what they’re looking for faster while boosting rankings and optimizing CPC and CAC. ### Easily Curate Featured Collections and Trend Campaigns Easily curate and launch promotions to help shoppers explore all the latest looks and trends. ### Sharpen Demand Forecasting Optimize assortment planning by uncovering new, actionable insights into product purchases and consumer preferences. --- # How Better Website Navigation Can Improve Your Customer Experience and Sales https://www.lily.ai/blog/how-better-website-navigation-can-improve-your-customer-experience-and-sales 2022-03-07 The average webpage visit lasts less than a minute. Whether you’re planning a new site or optimizing an existing one, here’s how to turn browsers into buyers with clear navigation. [image: How Better Website Navigation Can Improve Your Customer Experience and Sales] If you’re looking for an enlightening perspective on your online store, try looking at it for less than a minute. That’s the duration of the average webpage visit, and the view is very different from what the designer, developer or store owner sees. At a glance, important things disappear, mostly because they were designed by people who are immersed daily in the site’s architecture. Whether you’re planning a new site or optimizing an existing site, here’s how to turn browsers into buyers with clear, user-friendly website navigation. ## Best Website Navigation Practices for a New Retail Store Build If you’ve got the idea for a store but not the site, the prudent approach is to keep it simple at first and optimize as you grow. Most e-commerce platforms, such as Shopify, offer a proven and tested template that conforms to navigation best practices. These will allow you to get eyes on your products, but the ultimate goal is to appeal to repeat visitors. This means following these trusted site navigation techniques such as: - Setting a clear product hierarchy. Group your products into parent categories and subcategories to reduce decision fatigue for your customer. Don’t overdo it on the subcategories, however. According to the “three-click rule” favored by User Experience (UX) experts, it shouldn’t take more than three clicks to reach any inner page on your site from the home page. - Including a search bar! Seriously, it’s where most shoppers start, so make it prominent. Users typically give 23% of their attention to the site header, so that’s where your important features need to be, including the search bar, shopping cart icon and contact details. - Trimming your navigation menu. Save the top navigation menu for the essentials—ideally no more than seven tabs to limit decision paralysis. This is important for SEO purposes, too. Your page authority suffers in search if there is a perceived lack of focus in your navigation menu. - Using secondary navigation. It doesn’t have to be visible until the user hovers or clicks above, but a secondary navigation menu is a good place to include your social proof icons, current bestsellers and running promotions. Keep it updated, too! - Avoiding dropdown menus. The verdict is in: Dropdown menus are frustrating for customers, especially on mobile, and their content can’t always be found by search engine crawlers. Online fashion retailers, such as ASOS and Boohoo, get straight to the point with mega menus instead, where everything is visible (pictures, too) on one page. - ‍Filling your footer. The footer is too often treated as an afterthought in e-commerce, or used as somewhere to hide your legal disclaimers and other small print. That’s a wasted opportunity – the footer is premium real estate. Be sure to fill it with links to your important pages, contact details, and all the information about shipping, returns and so on that your customers crave. ## Optimizing an Existing Retail Website This is where we imagine most of this blog’s readers are. The user behavior flow in your Google analytics report is a vital resource for exposing bottlenecks, roadblocks or other clues that your website navigation needs tweaking. If in doubt, remove what’s not working, rename what’s underperforming and reposition what’s overlooked—and keep testing relentlessly! To turn a good site into a great store, try the following: - Rewrite labels. Replace any generic labels (e.g., products) with the terms that customers are actually searching for. You’re helping search engines, too. Shopify quotes the example of Harvard Business Services, which increased sales by more than 15% simply by rewriting labels. - Make search engaging. Add filters and facets to search (like Amazon does) so that users can refine their search from the outset, and drive traffic to your best performing product pages by adding context, such as how many products you have in each category. ‍ - Leave breadcrumbs. Let the user know where they are now (i.e., progress bar) and how to go back to a previous page, without having to input information again. This element of website navigation is a blend of sidebar and menus, and intelligent use of color and texture (e.g., highlighting) for better UX. ## These Navigation Errors Are Costing You Conversions It’s easy to spot a website where copy, images and design are not aligned. Here’s how to stand out for all the right reasons: - Keep it thorough, but succinct. Product descriptions should be detailed, but concise. They should be adding context to the images, not simply repeating what they show. - Focus on SEO. Headings and subheadings should be optimized for search, so short- and long-tail keywords are more important than quirky brand statements. - Build the right menu. Navigation menus should be consistent and standardized. Visitors expect to find them on the top left on every page. - Create clear CTAs. The number one rule is to never force your visitor to search for a checkout button on a product page. Calls to action (CTAs) should be prominent, placed in a variety of positions, and clear about what they lead to. Responsive page design is a must. Given that 73.1% of users will leave a website that is non-responsive, make sure your site is optimized for mobile and tablet, not just desktop. ## Getting Checkout Right The average cart abandonment rate at checkout is just under 70%. That’s a lot of effort wasted at the final hurdle. Create a frictionless, seamless checkout experience with the following: - Minimize distractions by removing the headers and footers that feature on other pages. - Use pre-filled form fields for faster checkout. - Allow a “continue shopping” or “modify order” function without losing saved information. - Show progress, letting customers know where they are in the shipping, payments and confirmation phases. - Highlight the product summary, shipping cost and product availability from the start. If customers reach the end of the process to find that their item is out of stock or too expensive to ship, they are unlikely to return. ## Improve Your Customers’ Experience with Lily AI Providing a slick, intuitive user experience through website navigation is one part of the challenge. Delivering a relevant shopping experience is the other. That’s where the powerful suite of data tools from Lily AI makes the difference, using AI and machine learning to show customers not just what they’re looking for, but also what they didn’t even know they wanted. --- # Fast-growing performance apparel brand: How a fast-growing performance apparel brand turned product data into a 7.3% revenue lift on Google Shopping https://www.lily.ai/case-studies/performance-apparel-google-shopping-revenue-lift A controlled 26-day Google Shopping test — with no change to bids or budget — drove a revenue lift confirmed causal with difference-in-differences analysis. - +7.3% — Revenue per product per day - +4.7% — Causal lift (difference-in-differences) - +1.3% — ROAS improvement - -5% — Cost per acquisition ($1.28 → $1.22) ## The shift: the feed was the ceiling, not the campaign The brand was already running a sophisticated Google Shopping program. Bids and budgets were tuned. But like most retailers, its product feed spoke the language of the business: a title, a short description, brand, and color. The attributes shoppers actually search on — fit, fabric, use case, occasion — were missing or buried in prose. The team wanted an honest answer to a specific question: could better product data, and nothing else, move revenue? Not a projection, not a demo. A tested result. ## The approach: a clean, controlled test Rather than promise a number, Lily AI ran a controlled experiment on Google Shopping. The catalog was split into a matched control group and a treatment group of comparable size. The control kept the brand's existing product content. The treatment group received Lily's enrichment: structured attributes, benefit-driven descriptions, consumer-language highlights, and search descriptors that reflect how people actually shop. Bids, budget, and campaign structure stayed identical across every group. The only variable was the product data. The test ran for 26 days. ## Why the method matters During the test, impressions and clicks fell across every group, including the control, by a similar amount. That was a market-wide effect, not the test. Because the control absorbed the same market conditions, a difference-in-differences analysis could separate what the enrichment actually caused from what would have happened anyway. This is the difference between watching a dashboard move and proving your work moved it. Demand that was already shifting affected every group equally, so it cancels out. What remained was the effect of the product data itself. ## What this means The campaign was never the constraint. The product data was. When the feed describes products the way shoppers and machines actually read them, the same ad spend produces more revenue, and the lift holds up under controlled measurement. The same enrichment layer that lifted Google Shopping here extends to Meta Ads, to organic and AI-driven discovery, and to onsite search. Enrich once, perform across every surface where people search and shop. --- # Global prestige beauty company: How a global prestige beauty company turned product content into cross-channel performance and AI-recommendation visibility https://www.lily.ai/case-studies/prestige-beauty-cross-channel-ai-visibility Controlled tests across a portfolio of makeup and skincare brands showed that enriching product content lifted revenue and efficiency on Google and Meta — and improved how often the brands surfaced in AI product recommendations. - -42% — Cost per purchase (Meta lift study) - +28% — ROAS (Meta lift study) - Double-digit — Google Shopping revenue lift - Stronger — AI-recommendation visibility ## The challenge Beauty shoppers do not search the way product catalogs are written. They ask for “a buildable cushion blush for sensitive skin,” not “face blush, paraben-free, peach.” That language gap costs visibility in Google Shopping auctions, on retailer sites, and increasingly in the AI-driven surfaces (ChatGPT, Gemini, AI Overviews) where shoppers now ask for recommendations directly. The company needed to close that gap across paid media, organic discovery, and the emerging AI recommendation layer, without rebuilding catalog operations or changing brand voice. ## The solution Lily Max, the agentic product intelligence engine, enriched the portfolio's product feeds with the consumer-language attributes, structured detail, and intent signals that ranking algorithms and LLMs actually use to match products to shoppers. What Lily optimized: - Product descriptions: attribute-dense, consumer-language copy that matches how shoppers actually search. - Structured attributes: comprehensive fields (color, finish, formulation, skin type, occasion, ingredients, claims) using Lily's beauty taxonomy. - Product highlights: fields surfacing key benefits and differentiators for ranking signals. - Content-quality scores: systematic improvement across the entire catalog in Google Merchant Center. ## Google Shopping: a controlled test Difference-in-differences A/B tests measured Lily-enriched feeds against matched control sets in Google Merchant Center, with downstream revenue tracked in the brands' own analytics. Across the portfolio, enriched content drove conversion-value and revenue lifts. Just as telling, enrichment improved traffic quality: in several cases the enriched feeds drew fewer but better-qualified clicks and produced stronger revenue per product. Richer content surfaces products to higher-intent shoppers, not just more of them. ## Meta: a conversion lift study A separate two-cell conversion lift study tested Lily-enriched product content against standard catalog content in always-on paid social. On the primary purchase objective, the enriched cell delivered 42% lower cost per purchase and 28% higher ROAS. The cross-channel pattern was consistent: better content quality compounds wherever ranking and bidding algorithms read product data. ## The AI visibility dividend The same enriched content is what LLMs read when they generate product recommendations. Better content quality drives stronger performance and engagement signals, and those are the signals AI systems ingest. After enrichment, the portfolio's brands surfaced more often in AI-generated product recommendations in their categories. Lily improves the readiness and quality of the data these systems read. It does not control final placement or citation, and it makes no such promise. What it does is make a brand the easiest, best-described option for the machine to choose. ## The takeaway Lily Max does not just optimize feeds. It shapes how algorithms and AI see your brand. For this beauty company, that meant revenue and efficiency lifts in paid channels and stronger visibility in the AI recommendation surfaces that increasingly mediate beauty discovery. Product content quality compounds across every surface where shoppers, and the algorithms guiding them, make decisions. --- # Global home furnishings retailer: How a global home furnishings retailer grew Google Shopping revenue 5.7%, while clicks went down https://www.lily.ai/case-studies/home-furnishings-google-shopping-revenue-lift A controlled 28-day A/B test in Google Merchant Center showed that enriching product data — with no change to bids or budget — drove more revenue from better-qualified traffic, even as clicks fell. - +5.7% — Google Shopping revenue per product - +3.2% — Impressions - -1.3% — Clicks (fewer, higher-intent) - 28 days — Matched-control A/B test ## The shift: more spend was not the answer The retailer already ran a large, mature Google Shopping and paid search program. The obvious levers — bids, budget, campaign structure — were already well tuned. But like most retailers, its product feed was written in the language of the business rather than the language shoppers use. The question was whether better product data alone, with no change to spend, could still move revenue at this scale. ## The approach: a clean, controlled test Lily AI ran a controlled A/B test in Google Merchant Center. The catalog was split into a matched treatment group and a control group. The control kept the retailer's existing product content. The treatment group received Lily's enrichment: richer, consumer-language product data and structured attributes that Google's ranking and matching systems actually read. Bids, budget, and campaign structure stayed the same across both groups. The only variable was the product data. The test ran for 28 days. ## What the results say Revenue rose while clicks fell. That is the interesting part: the enriched content was surfacing products to more qualified, higher-intent shoppers, not just more of them — fewer wasted clicks, more revenue per click, better efficiency on the same spend. For a retailer of this size, the lift ran above the average Lily sees for comparably large catalogs. ## What this means At this scale, a mid-single-digit revenue lift from product data alone, with spend held flat, is a meaningful result, and the efficiency gain compounds over time. The campaign was already optimized. The product data was the lever that still had room to move. The same enrichment layer extends beyond Google Shopping to Meta Ads and to organic and AI-driven discovery. Enrich once, perform across every surface where people search and shop. --- # A.L.C.: How A.L.C. boosted full-catalog product discovery on Google https://www.lily.ai/case-studies/alc-full-catalog-discovery-on-google A.L.C. enriched its product attributes with Lily AI to expand discoverability across its full catalog, on-site and on Google, driving a major lift in non-branded search and net sales. - +129% — Lift in net sales - +81% — Lift in net items sold - 88% — Of impressions from non-brand search - +13% — Lift in GMC clicks ## The Challenge In 2022, A.L.C., an advanced contemporary brand primarily known for dresses and some key iconic styles, wanted to improve the discoverability of its broader product range to drive growth. Customers primarily found a limited selection of key dress styles through organic and paid search, hindering the brand’s efforts to expand into other categories like sweaters and trousers. That’s when Hilary Peterson, the SVP of eCommerce and Digital Advertising at A.L.C., reached out to Lily AI. ## The Solution: Lily AI Product Content Optimization A.L.C. partnered with Lily AI to enhance its product attributes and improve search performance, starting first with on-site search and then expanding into Google, both paid and organic. > “Pretty much right away in 2022, we added the attributes to our product catalog, and we saw a huge shift and a huge lift in our non-branded search terms,” said Hilary. “Non-brand drove 88% of impressions and 40% of clicks, and the volume of products and people coming to us expanded dramatically. And we actually saw this over time, trickle down to building authority in other categories.” By implementing Lily AI’s product attribute and content enrichment, A.L.C. was able to achieve the following: – Improved Organic Search: Expand non-branded search terms to drive more traffic to a wider range of products. – Optimized Paid Search: Boost paid search performance across product visibility, web traffic, and sales. – Adapted Faster to Market Trends: Improved ability to quickly respond to seasonal changes, trends, and cultural moments to ensure product content remains dynamic and relevant. – Enhanced Cross-Functional Collaboration: Able to now foster a more collaborative relationship between marketing and merchandising teams by leveraging Lily AI’s data and insights. ## The Results The implementation of Lily AI’s product content optimization solution delivered early wins in 2022 and continued to drive impressive results in testing the latest release of Lily AI’s Product Content Optimization platform. In the graph below, A.L.C.’s agency, January Digital, charted the trendline of nonbrand Google impressions from 2021-2024, with a clear increase around the time of the Lily AI implementation. Specifically, since working with Lily AI as well as in recent testing, A.L.C. has seen significant positive results: – Increased Non-Branded Search: A substantial lift in non-branded search terms, driving 88% of impressions and 40% of clicks, reaching new customers. – Boosted Sales for More Items: Saw a 129% lift in net sales, with an 81% lift in net items sold and at a higher AOV per unit, boosting discoverability for a wider assortment of higher margin products. – GMC and Paid Search Optimization: A 13% lift in clicks alone by adding product highlights as another data enrichment opportunity within GMC (Google Merchant Center). – Enhanced Team Collaboration: A more collaborative relationship between A.L.C.’s marketing and merchandising teams, leveraging Lily AI’s data and insights for copywriting and product launch strategies. > “We saw a tremendous lift in the products that had Lily AI’s attributes attached to it across both volume of products sold and then also the revenue attached to it. So not only did we sell more units, we also sold more at a higher AOV, which was fantastic.” ## Conclusion A.L.C.’s partnership with Lily AI has enabled the brand to overcome challenges in product discovery, improve search performance, and foster a data-driven and collaborative approach to marketing and merchandising. > “Now, we use the [Lily AI] insights and data to give to our marketing team. And so they use that to help write the copy based on trends you’re seeing. Same with our merchandising team. It’s become much more collaborative and cross-functional.” By embracing innovation and continually testing new solutions, A.L.C. has achieved significant results and is well-positioned for future growth. > We saw a tremendous lift in the products that had Lily AI's attributes attached to it. Not only did we sell more units, we sold more at a higher AOV. — Hilary Peterson, SVP of eCommerce & Digital Advertising, A.L.C. --- # thredUP: thredUP increases sell-through by 15% with enriched product data https://www.lily.ai/case-studies/thredup-sell-through-15-percent AI-driven deep tagging from Lily AI gave thredUP shoppers two to three times as many ways to search and discover its single-SKU inventory, lifting sell-through by 15%. - +15% — Lift in sell-through - 275k+ — Images tagged per week - +2.8% — Conversion lift on iOS - 2-3x — More searchable attributes ## The Problem: Millions of unique items, thousands of ways to look for them With more than 100,000 items arriving every day from sellers’ closets, thredUP must categorize and tag products at tremendous scale. Its inventory includes more than 50,000 brands and hundreds of categories, each with dozens of attribute groups—any of which could lead a buyer to the item she wants. thredUP associates, working in distribution centers across the United States, inspect each item that arrives from a seller’s closet and create a unique product description for it. Before thredUP began working with Lily AI, the associates were also responsible for tagging inventory items with all of the attributes that the site uses for buyer navigation, searches, facets, and filters. This process could be time-consuming and required a robust QA process to ensure a good customer experience. Chris Homer, co-founder and CTO at thredUP, leads a team of software engineering and data science experts who work together to eliminate the friction from consignment selling. He knows that search and navigation are critical to his customers. However, he aims to keep internal resources focused on technology and algorithms that are unique to thredUP’s business. “We need to invest in our pricing algorithms, scoring of seller merchandise, and garment routing in the distribution center—critical profitability drivers that are unique to us,” Homer says. “When it comes to personalization and recommendations, which involve customer behavior, products, and affinities that are relatively consistent across retailers, we want to find the best partners.” ## The Solution: AI-driven deep tagging While the company had conducted pilots of technology-based tagging solutions, it had not yet found a solution that met its goals for accuracy or reliability until it began working with Lily AI. Homer’s team evaluated Lily AI Product Intelligence through a pilot with women’s tops, a product category containing a wide variety of styles and used for diverse occasions. They wanted to understand the quality that Lily AI algorithms could deliver as well as how the solution would improve the efficiency of human taggers. The company’ early split test with Lily AI showed a sell-through lift of four to 15 percent. Based on these results, Homer’s team made the decision to expand Lily AI’s deep tagging across all of its product categories. ## The Results: More ways to guide customers to the perfect choice With Lily AI Product Intelligence, thredUP customers now have two to three times as many attributes to search and explore as they look for the perfect outfit. On average, Lily AI tags about 275,000 images each week. This has resulted in a 15% lift in sell-through; a 2%+ conversion rate lift for customers who’ve purchased >2 times, and a 2.8%+ conversion rate lift on iOS. Favorites per user increased by 2.1% as well, with favoriting being an important feature on thredUP due to the many single SKUs on offer. ## Improved operational efficiency and scale Operational efficiency has been another important benefit for thredUP. Lily AI Product Intelligence has reduced the training requirements for thredUP associates and reduced QA for newly added products, so they reach customers faster. “It’s like removing extra weight from the system, and that makes a bigger and bigger difference as we scale up,” Homer adds. Furthermore, thredUP is able to use extended attribute data from Lily AI in its pricing algorithms, which are critical to sell-through. The algorithms now have more signals to assess the value of and expected demand for a particular garment that a seller has consigned. > “It all needs to start with the customer. The context of what they are trying to accomplish — their progress in refreshing their closet or finding a specific piece for an event — does not come from segmentation alone. Lily AI’s customer intent platform understands what our customer is trying to do and offers the right recommendations to them.” — Chris Homer, Co-Founder and COO, thredUP > When it comes to personalization and recommendations, we want to find the best partners. — Chris Homer, Co-founder & CTO, thredUP --- # Global luxury retailer: A global luxury retailer increases relevant results for descriptive searches by up to 30x https://www.lily.ai/case-studies/luxury-retailer-descriptive-search-30x Over half of this retailer's on-site searches were attribute-heavy, long-tail queries its manual attribution couldn't serve. Lily AI's consumer-centric attributes increased relevant results by up to 30x. - Up to 30x — More relevant results for descriptive searches - +3.5% — Increase in online order conversions - 75k+ — Products tagged in under two weeks ## The Problem In 2020, the luxury retailer began a deep look into some of its key e- commerce business metrics, and had a hypothesis that conversion from online searches could be improved to meet internal standards and outpace industry norms. In setting out to understand areas for improvement, the team sought to understand the precise searches for all site visitors. They found that while many searches were focused on either distinct brand names or on broad categories (i.e. “dresses” or “shirts”), over 50% of searches were attribute-heavy, long-tail searches that combined materials, trends, occasions and other customer-centric attributes. An example from Q4 2020 found that shoppers were often searching for “sequin” in the retailer’s on-site search engine, and were returned between 10 to 20 sequined dresses; and yet the company knew they had over 200 sequined dresses in their assortment. When they analyzed why this was happening, they came to realize that their manual, human-based, merchant- driven attribution was a bottleneck in terms of consistency, speed and accuracy. The company wanted to reach these high-intent customers, and knew that Lily Al could help. ## The Solution Understanding that getting this right was a huge sales opportunity, the Company entered into a proof of concept with Lily Al that took a specific set of products – initially, dresses — and, using historical data sets and Al-based computer vision, appended Lily Al’s extensive customer-oriented product attribution to the retailer’s existing product data. Measuring initial success here, the relationship expanded across apparel, shoes, handbags, jewelry, and accessories, with over 75,000 products tagged in less than two weeks. In an interview, the retailer’s SVP had this to say: “Then we saw something rather remarkable. There were two metrics that were really interesting. The first one was, of the many people that I show these better attributed results to, how many people click into a product page? That increase was kind of mid-single digits. Super interesting to me, because again, these are really high-intent customers, who are now finding a lot more of interest for them to click into. Now I’ve kept them on the website longer, and I am getting them to a product page; so further down my funnel. Even more interesting was the metric of: are they actually converting on those products, and they were converting even more. I saw that if I could show them things that were more relevant, they’re going to click on that; they’re going to spend more time on my site exploring, and they’re much more likely to convert. So this was hugely interesting, because it told me, basically, there’s a lot of lift for my overall business for any product where I could attribute them in this more reliable way. I’m engaging customers by showing them more relevant products, and I’m helping them complete their shopping journey, which is good for them and good for me.” Having seen exceptionally strong results in our proof of concept, the retailer extended the relationship with Lily Al to leverage Lily Al’s granular and consumer-centric product attributes to enhance site search across apparel, shoes, handbags, jewelry and more. ## The Results By using Lily Al’s consumer-focused tags to augment their on-site search, the retailer increased the relevant results for descriptive searches by up to 30x. This translated to incremental revenue for the Company, which led to an expansion of Lily Al product attribution across the Company’s diverse, varied, and rapidly-growing product catalog. Lily Al’s product data enrichment engine, trained on over 1 billion data points, delivers product attributes with unparalleled detail and accuracy, enabling an intuitive digital experience. The solution also enforces consistency across the entire product catalog, which is extremely valuable for large retailers such as this one, and then pushes product data to the entire e-commerce stack, creating a universal, customer-centric language that can be used to enhance search, product recommendations, SEO/SEM and demand forecasting. --- # Large multi-brand retailer: A large multi-brand retailer boosts product discovery, projecting a $22M revenue increase https://www.lily.ai/case-studies/multi-brand-retailer-22m-product-discovery Richer Lily AI attribution lifted on-site discoverability 8.5% and search-driven demand 3.3% for a large multi-brand specialty retailer, projecting to a $22M revenue increase across brands. - +8.5% — Increase in discoverability - +3.3% — Increase in search-driven demand - $22M — Projected revenue increase - +33% — More product attributes assigned In 2021, a large, multi-brand speciality retailer had been encountering multiple challenges in ensuring that their extensive catalog of clothing and apparel was being discovered by shoppers on their online properties – most specifically via on-site search. The retailer turned to Lily Al and its product data enrichment platform for help. ## The Problem Given its challenges, the retailer saw a need for better product data as a huge value unlock for on-site search, one that might lead to an increase in the number of relevant products shown to a customer, thereby improving discoverability and ultimately demand for that product. A subset of CC’s were provided to Lily Al to add product attribute tagging to, with a control vs. product that was not tagged using Lily Al. The primary KPI this retailer sought to measure was demand and how it was impacted based on product views, product discoverability and search intent. The retailer was looking to improve their core customer journey with deep and consistent product attribution, from Lily Al, and to show an 8-digit lift in topline revenue. The approach needed to be a low lift integration and implementation on the brands side as well, which Lily Al was able to provide. Lily Al’s output for this search test was then deployed into Bloomreach for full implementation. ## The Results The multi-brand retailer saw very rapid and very measurable results in on- site search using the Lily Al platform. Product attributes assigned by using Lily Al increased by an average of 33% (and from 8 attributes to 24 in 1 particular product example). The increase in search-driven demand was determined to be all net additional topline revenue, and was not displacing sales from other avenues. This led to an 8.5% increase in discoverability, measured by the percentage of PDP views that came directly from search — thus increasing the likelihood of a relevant product to be shown to a customer based on their search query. This in turn resulted in a 3.3% increase in demand driven purely from search, projecting to a $22M increase in revenue across all brands. --- # Large multi-brand retailer: Lily AI recommendations increase RPV by 0.65%, driving $20-30M in incremental demand https://www.lily.ai/case-studies/multi-brand-retailer-rpv-recommendations Layering Lily AI's granular attributes onto an existing Certona recommendation engine produced more relevant recommendations, lifting revenue per visit and order size for a $20-30M incremental-demand projection. - +0.65% — Increase in revenue per visit - +0.6% — Increase in average order size - +0.62% — Increase in overall demand - $20-30M — Projected incremental demand The challenge of matching shoppers to the products they actually want via recommendations is a constant across the retail e-commerce landscape. A large, multi-brand specialty retailer came to Lily AI with a product recommendation engine that they had been working with, concerned that this engine was not able to take advantage of granular product attribute data to ensure that consumers were being presented with items that truly matched their personal style and preferences. This retailer turned to Lily AI and its product Attributes platform for help. ## Highlights - 0.6% increase in AOS - 0.65% increase in RPV - 0.62% increase in overall demand - $20-30M increase in annual revenue ## The Problem The large multi-brand retailer had been looking to offer the same benefits of in-store shopping that might come from a stylist/sales associate showing a customer similar product recommendations to what they are looking for, yet in an online platform. For instance, a customer might be looking at an eyelet top, and wants to see other options that are similar. The recommendation engine the retailer had been using could not identify eyelets or other details of the viewed product, but they surmised that Lily AI could. In the initial proof-of-value test, Lily AI was able to provide more granular product information, which in turn provided more context and relevant product recommendations that could then be input into the existing recommendation engine. This allowed the retailer to develop a more hyper-personalized product set, thereby increasing the number of relevant products shown to a customer. Lily AI’s output for the recommendations test was deployed into Certona, the company’s product recommendation engine. KPIs that this retailer sought to measure included RPV (revenue per visit), AOS (average order size) and overall demand—all of which were hypothesized to increase with an increase in relevant product recommendations and views. One interesting discovery during the retailer’s proof-of-value tests was that visits, orders, CVR, and AUR (average unite retail) all remained constant, which indicated that the basket size increase was what was driving the demand. Customers were adding more “quality” products to their shopping bags, based on seeing more relevant product recommendations. ## Results With Epsilon Research finding that 80% of consumers are more likely to make a purchase when brands offer personalized experiences, delivering robust product recommendations that convert is core to any retailer or brand’s e-commerce strategy. By layering on granular product attribution data to their existing recommendations solution, Certona, this retailer saw an overall increase in RPV, AOS, and overall demand thanks to Lily AI, driven by a substantial increase in product views. They saw an increase in AOS of 0.6%, a 0.65% increase in RPV, and a total increase in overall demand of 0.62%, for this retailer’s size, this projects to an additional $20-30M in incremental demand. --- # Multi-brand apparel & accessories retailer: Lily AI demand forecasting drives 7-8 digit revenue lift with accuracy at scale https://www.lily.ai/case-studies/demand-forecasting-accuracy-at-scale Lily AI's visual-similarity models replaced manual proxy-product guesswork, cutting a retailer's forecasting cycle from three months to one and projecting a $7-48M topline revenue lift. - $7-48M — Projected topline revenue impact - 3 mo → 1 mo — Faster forecasting cycle - 34k+ — Products forecast each year --- # Luxury home brand: How a luxury home brand boosted sales from Shopify site search and Google search https://www.lily.ai/case-studies/luxury-home-brand-shopify-and-google A top luxury home furnishing brand enriched its Shopify product content and GMC feed with Lily AI, lifting on-site revenue per visit, Google impressions, and ad clicks, with a ROAS boost two weeks after going live. - +3.2% — Lift in on-site ARPV - +8% — Impressions on Google - +3% — Increase in ad clicks - 2 weeks — To a measurable ROAS boost ## Challenge Part 1: Shopify Site Search A top luxury home furnishing brand needed to transform its digital shopping experience and accelerate on-site conversion rate optimization (CRO) initiatives. To do so efficiently and effectively, the brand needed a solution that would seamlessly integrate with the Shopify platform and Shopify on-site search. The brand suspected that sub-optimal product attributes and metadata were impacting e-commerce conversion rate performance. Specifically, they wanted to address issues within site search, including null search results and irrelevant products being returned, which they knew was leading to a poor customer experience and leaving sales on the accent table. So the brand turned to Lily AI for help to address their e-commerce data issues first, yet with an eye for expansion to Google next. ## The Solution: Lily AI Product Content Optimization The brand partnered with Lily AI to first enhance its product content in Shopify, specifically, non-consumer-facing product details, attributes, and metadata. This included auditing and verifying product data accuracy, and then adding missing micro-details (fixtures, arm style, wood tone) as well as more subjective descriptors (home setting, season or holiday, decor style) and natural language synonyms. After increasing Average Revenue Per Visit (ARPV) by 3.2% through on-site search improvements, this luxury home furniture brand turned its eye to optimizing Google Search performance to drive more traffic to its conversion-optimized website. ## Challenge Part 2: Expansion to Google SEM and SEO Through the Lily AI Google Merchant Center (GMC) integration, the brand was able to automatically enrich the product data being fed to Google through an easy supplemental GMC feed. Optimized with Google-friendly, natural language attributes and new product details, the brand’s performance marketing team was able to increase brand and product visibility on Google, specifically premium Product Listing Ads (PLAs) and Product Listings (PLs) placements. ## Conclusion: Compounded Success The optimized feed, compounded with optimized product content on the PDP, helped boost Google Quality Scores and Ad Rank, which further drove more impressions and clicks without increasing media costs. The result: incredible speed to impact with a significant improvement in ROAS after only 2 weeks of going live with Lily on Google. Combined, the brand saw measurable wins across the board: ## +3.2% Lift in On-site ARPV +8% Impressions on Google +3% Increase in Ad Clicks ↑ROAS Boost After 2 weeks --- # Tapestry: Tapestry boosts traffic and sales from Google Search with Lily AI https://www.lily.ai/case-studies/tapestry-google-search-discovery At CommerceNext 2024, Tapestry (parent of Coach) and Lily AI showed how enriching product attributes with customer-centric language drove high-single and double-digit lifts across SEM, SEO, and Site Search. - Double-digit — SEO lifts: sales & PDP views per SKU - Double-digit — SEM lifts: sales & PDP views per SKU - High-single+ — Conversions & avg. sales per view At the 2024 CommerceNext, the Lily AI team was on the scene and helping brands and retailers bring humanity to shopping. In fact, our President Ahmed Naiem was joined on stage by Tapestry’s Vice President of Global Digital Product and Omnichannel Innovation, Mandeep Bhatia, to share how Lily AI is helping Tapestry do just that. Their session “Results Your CFO Can Believe In: How Tapestry Is Winning With AI” illustrated two big problems in retail today, and the financial impact of solving it: - Customer speak is vast: It’s trends. It’s styles. It’s occasions and micro-occasions. It’s vamp coverage and fabric quality, plus synonyms for each of these highly-detailed and often subjective attributes that matter since we all have our unique preferences and ways of describing what we want. - Traditional “merchant speak” is narrow: It’s fabric. It’s color. It’s hardware. Most products are described today using a very common yet narrow taxonomy of product attributes that only begin to scratch the surface for how we discover and shop for new items. The widening gap between the language of the customer and the language of brands and retailers can be a challenge—or an opportunity. In their session, Ahmed and Mandeep went beyond ROI and ROAS to explore how Tapestry is approaching their investment in Lily AI to drive both significant and immediate impact. Specifically, Tapestry partnered with Lily AI to maximize product discovery via Google Search and Site Search. They are also planning to expand the partnership into e-commerce to further personalize customer shopping experiences across all of Tapestry’s brands. ## How Lily AI Optimizes Search for Tapestry Ahmed showed how Lily AI enriches product attributes with customer-centric language to connect people to products. He used the iconic Y2K-inspired Tabby Shoulder Bag by Coach as an example. First, he showed off the product attributes that were initially associated with the bag. These attributes and images were fairly standard. From there, Lily used a combination of generative AI, computer vision, natural language processing, machine learning, and deep learning which all analyzed the item and suggested more thorough, customer-centric attributes. Dozens more attributes, attribute synonyms, and macro- and micro-fashion trends were then included in the handbag’s product attributes, supercharging its discoverability when distributed to Google Merchant Center and their e-Commerce platform The results? Lily AI boosted Google Search performance for Tapestry! Now, for example, products like this version of the Tabby bag are much more easily found on both Google Search and on Coach’s website via site search. When a consumer searches for something like “glamorous adjustable crossbody for luncheon” they’ll easily find relevant product results that lead to a sale. The Lily AI-Tapestry collaboration drove high-single and double-digit lifts across SEM, SEO, and Site Search key metrics: - ⬆️ Conversions - ⬆️ Average Sales per View - ⬆️⬆️ Avg. sales per SKU (SEO) - ⬆️⬆️ Avg. PDP views per SKU (SEO) - ⬆️⬆️ Avg. sales per SKU (SEM) - ⬆️⬆️ Avg. PDP views per SKU (SEM) If you’re ready to improve the effectiveness of both organic and paid advertising strategies on Google Search and beyond, let’s chat! --- # Rue Gilt Groupe: From experimentation to full AI vibes at Rue Gilt Groupe https://www.lily.ai/case-studies/rue-gilt-groupe-full-ai-vibes Rue Gilt Groupe SVP Product Josh Platt and Lily AI CEO Purva Gupta at CommerceNext 2025 on putting AI to work across e-commerce and advertising. ## Lily AI Returns to the CommerceNext Stage for the Third Year In 2023, Lily AI invited Danielle Schmelkin, CIO at J.Crew, to share their journey in enhancing on-site search to increase relevance and minimize null results. In 2024, Lily AI invited Mandeep Bhatia, VP, Global Digital Product & Omnichannel Innovation at Tapestry, to share their experiences optimizing organic and paid Google shopping ads to boost ad impressions and clicks. Now at CommerceNext 2025, Lily AI is excited to welcome Josh Platt, SVP Product at Rue Gilt Groupe, to share their story of testing and innovating with Lily AI across e-commerce and advertising. ## From Experimentation to Full Vibes Rue Gilt Groupe is as much a retailer as it is a technology company, blending proprietary in-house innovation with best-in-class industry solutions to deliver dynamic, personalized shopping experiences. By embracing AI across its operations, Rue Gilt Groupe is making search and shopping more relevant and enjoyable for its customers, driving higher performance and productivity. In this session, Josh, an experienced Product, AI & UX leader with an extensive history building ecommerce and omnichannel experiences with an emphasis on AI-driven innovation, will join Lily AI Co-Founder and CEO, Purva Gupta, to discuss the secrets to their successful partnership and how to vibe with AI in advertising and e-commerce. We hope to see you at our session on the first day, Tuesday, June 24th at 3:45pm in the AI for Marketing Track, Breakout Session #3! --- # Merrell: From SEM to GEO: how Merrell connects with Gen Z and millennials https://www.lily.ai/case-studies/merrell-sem-to-geo Merrell's VP of Digital Marketing on how the brand is moving from search-engine to generative-engine optimization to reach Gen Z and millennials. ## Key Takeaways: - Consumers are fast adopting AI-powered shopping, with Millennials and Gen Z leading the way. - Search queries are becoming longer and conversational since AI can now understand intent. - Merrell is investing in Product Content Optimization to boost product discoverability. In our recent eMarketer webinar with Jane Smith, VP of Digital Marketing at Merrell, Jane shed light on how Merrell is navigating and thriving in the shifting landscape of consumer discovery that is increasingly being influenced and impact by AI. As just one example, their customers are now searching using detailed, conversational queries averaging 23 words. Jane highlighted Merrell’s strategy to align product details with consumer search trends, emphasizing real-world context such as comfort, style, and usability in various environments. Jane also emphasized the importance of ongoing optimization and the seamless integration of Lily AI’s insights into Merrell’s existing systems to enhance consumer discovery. The inspired some fantastic discussion as well as lively Q&A from the audience. With that, here are all the answers to your burning questions! ## Q: What is GEO? A: It sure is confusing when one web of concepts is known by so many distinct, yet related acronyms! Let’s define GEO: An emerging new discipline that enhances digital content to be found and favored by AI-powered search engines. The retailer’s goal is similar to Search Engine Optimization (SEO), which involves increasing product visibility organically and driving more traffic to an e-commerce website. Other related yet distinct terms and concepts include: – AEO (Answer Engine Optimization) – GSO (Generative Search Optimization) – AISO (AI Search Optimization) – AIO (AI Overviews Optimization) – LLMO (Large Language Model Optimization) ## Q: What are the potential pitfalls when adapting to AI-powered search strategies? A: Brands should avoid treating content optimization as a singular event. Instead, it must be viewed as a dynamic, ongoing effort continuously informed by new consumer trends and behaviors. Outdated or stale content also loses effectiveness in search engines and AI-driven environments. This is where AI helps to scale the work. Content optimization is an always-on activity, whereby the AI constantly observes not only new consumer search queries, but also the performance of the content itself, and then dynamically adjusts the existing product content accordingly to maximize relevance and results. Product content “freshness” and relevance are both key! ## Q: How quickly will consumers transition from traditional search to AI search? A: Retailers may feel justified in adopting a “wait and see” approach to investing in GEO search, citing slow consumer uptake, but that is simply not true. This moment is fundamentally different from prior innovation moments in advertising history. The momentum behind GEO is being driven by end consumers themselves. Big Tech’s alpha and beta test environments for GEO-powered search and discovery have been met with enthusiasm, with people continuing to engage even before the final, more refined and accurate versions are released. This organic, consumer-led adoption stands in stark contrast to prior innovations, such as retail media, as an example where industry investment outpaced shopper awareness or demand. The surge in GEO usage signals that consumers are actively shaping this market, not just passively responding to industry initiatives. To put this into numbers, in Lily AI’s consumer shopping survey conducted January 2025, 40% of consumersalready reported using AI-powered search specifically for shopping. While younger generations are leading in AI adoption, with 41% of Millennials and 45% of Gen Z reporting utlizing AI for shopping, Gen X adoption wasn’t far behind at 32%! Waiting on the sidelines risks missing a rare inflection point where consumer behavior (and not industry hype) is setting the pace for innovation and adoption. ## Q: What technical aspects are often overlooked yet critical for GEO discoverability? A: Traditional technical SEO included a whole checklist to maximize core web vitals, page speed, mobile optimization, etc. In the world of SEO and GEO, the importance of technical content is emerging as a critical pillar to help eCommerce sites communicate with AI through “machine speak”. This includes structured data such as Schema, Meta Titles and Descriptions, Alt Text, and more. This is content that consumers can sometimes see, and sometimes not, yet is fully optimized for search engine and AI indexability and crawlability. Oddly enough, one of the most overlooked technical aspects of GEO might be the role of humans in initial data creation and accuracy. As the old saying goes, garbage-in, garbage-out. In our work, we’ve seen a lot of wildly inaccurate product data which needs to be analyzed, cleansed, and corrected before the real content optimization work can begin. We’ve seen items with the incorrect country of origin, completely incorrect fabric content, mythical details, and features. It’s a problem. Last but not least, it can’t be stressed enough that we are living in a period where we must adapt and test while the inventions themselves are still being developed. So, as an example, in the old days of traditional SEO, it was (and still is) considered unethical to engage in Black-Hat SEO techniques such as keyword stuffing or hidden text and links. Now, new companies are emerging to create ghost versions of eCommerce websites, specifically designed to NOT be seen by humans but to be the source for AI agents. Think of it as a digital twin of your site. This may be a good short-term technological solution, yet we will see if this approach stands the test of time in terms of building consumer trust and operating with transparency, ethics, and scale. ## Q: How can brands optimize user-generated content (UGC) effectively for GEO and AEO? A: To optimize reviews and user-generated content for GEO and AEO, brands need to make this content easy for AI systems to understand and surface. This means structuring reviews with clear attributes and keywords, utilizing schema markup, and ensuring that images and videos have accompanying descriptive text. It’s also important to analyze and highlight the real language customers use so AI can match consumer queries with authentic experiences and insights. Finally, brands should measure which types of UGC drive the most discovery and conversions through GEO channels, and continually refine their approach. In short, treat UGC as a strategic asset: organize it, enrich it, and let it power your brand’s visibility in the new world of AI-driven search. ## Q: With voice and visual search becoming prominent, how will Product Content Optimization (PCO) evolve? A: The good thing about PCO is that while it is grounded in text-based data, voice, image, and video assets are able to be converted to text for context and intent understanding. PCO can handle the nuances of multimodal search and discovery because it understands natural language patterns and intent-driven inquiries. Additionally, it can handle things such as multimodal alignment, cross-modal consistency, contextual metadata, and composite indexing, all of which are critical, especially in visual search scenarios. While text remains a foundational layer, due to GEO’s fusion of modalities, PCO is more of a synchronized system where every asset, whether spoken, written, or visual, reinforces the others through AI-native contextual signals. Brands that utilize PCO to optimize for this interplay can dominate in voice and visual search environments as much as they can in text-based ones. ## Q: What new metrics are necessary for measuring the ROI of GEO initiatives? A: To be sure, we need a new measurement framework for GEO’s ROI, combining AI-centric metrics with refined traditional KPIs to capture this new, unique value proposition. Here at Lily, we like to say, you need to be found in order to be sold. As such, share of voice around brand and product mentions will be critical as well as new, core GEO-specific metrics that are emerging now. ## In Summary: Key to Merrell’s strategy has been the implementation of Product Content Optimization (PCO) with Lily AI, which ensures product details and descriptions better align with the way consumers naturally speak, shop, and search. --- # Bloomingdale's: How Bloomingdale's makes site search convert with enriched product data https://www.lily.ai/case-studies/bloomingdales-site-search-convert Bloomingdale's on lifting on-site search conversion with richer product attributes, so search results match the words shoppers actually use. --- # Tapestry: “AI is here, now what?” with Tapestry and Lily AI https://www.lily.ai/case-studies/tapestry-ai-is-here-now-what Tapestry and Lily AI on turning AI ambition into measurable retail results, and on where AI-enriched product data is already paying off in search. --- # J.Crew: How J.Crew uses AI to delight customers and power sales growth https://www.lily.ai/case-studies/jcrew-ai-delight-customers How J.Crew uses AI and product intelligence to delight customers, help shoppers discover the right products, and power sales growth. --- # Frequently asked questions https://www.lily.ai/faqs ## What Lily offers **What is Lily AI?** Lily AI is the company behind Lily Max, the product intelligence platform for AI commerce. Founded in 2015, Lily AI makes retail and brand catalogs legible to the search engines, ad platforms, and AI agents that decide what shoppers see. **What is agentic product intelligence?** Agentic product intelligence is structured, machine-readable product data built so AI agents and answer engines can understand, compare, and recommend products. Lily Max generates and maintains that layer so your catalog performs wherever AI does the discovering and the choosing. **How does Lily Max work?** Lily Max runs a continuous loop. It detects gaps in your product data, generates enrichment in consumer language, launches a controlled experiment, measures the lift against a holdout, and scales what works across your full catalog. **Who is Lily Max for?** Lily Max is built for retailers and brands that depend on product discovery, especially performance marketers and feed managers. It works for catalogs across fashion, footwear, beauty, skincare, and home, from growing brands to global enterprises. **Which channels and surfaces does Lily Max support?** Lily Max supports four surfaces: Google Ads, Meta Ads, AI discovery and agentic commerce, and onsite search and merchandising. On paid surfaces it improves measured performance. On AI surfaces it improves how readable and recommendable your products are. **What does Lily do?** Lily is an AI platform for retail media and agentic commerce. We connect product data, shopper intent, and paid channels so brands and retailers can grow profitably, across Google Ads, Meta Ads, onsite placements, and emerging AI discovery surfaces. **Which channels does Lily support?** Google Ads (Shopping, Performance Max, and Search), Meta Ads (Advantage+, catalog, and prospecting), onsite search and recommendations, plus AI discovery across agentic shopping surfaces and answer engines. **Is Lily a content generator or a feed manager?** Neither. Content is an output and feeds are a distribution layer. Lily Max improves the underlying product intelligence those outputs carry, and proves which improvements move performance before deploying them. ## Pricing and plans **How much does Lily Max cost?** Pricing is based on your catalog size and the surfaces you choose, structured as credits across Google Ads, Meta Ads, AI discovery and agentic, and onsite. Book a demo for pricing scoped to your catalog. **How is Lily priced?** Pricing is tailored to your catalog size, channels, and ad spend. We offer plans for growing brands through to enterprise retailers and marketplaces. Book a demo for a quote tailored to your business. **Is there a free trial?** We typically start with a scoped pilot so you can see measurable impact on your own catalog before committing. Talk to us about what a pilot looks like for your team. ## How Lily compares to other tools **How is Lily different from a feed management tool or an agency?** Feed tools move your catalog data and agencies manage your campaigns, but neither improves what the data actually says. Lily Max enriches the product attributes and language those systems depend on, then proves the lift with controlled tests. **How is Lily different from AI visibility tools?** Visibility tools tell you where you appear. Lily Max improves the inputs that determine whether AI systems surface your products at all, and ties those improvements back to revenue, not just a score. **How is Lily different from agentic-commerce-only (ACO) tools?** ACO optimizes agentic surfaces. Lily Max applies the same operating discipline across agentic commerce plus Google, Meta, organic, and onsite, so improvements compound instead of fragmenting per channel. ## Data collection, coverage, and accuracy **What data does Lily need to get started?** Lily Max works from your existing product feed and catalog data, the same titles, attributes, and descriptions your feed and PIM already hold. No replatforming or new data pipeline is required to begin. **How does Lily keep product intelligence accurate?** Lily Max treats every enrichment as a hypothesis: it tests which signals lift performance, deploys what wins, and feeds the outcome back into the model so accuracy improves with every cycle. **Do I keep control of my ad accounts and data?** Yes. Lily operates inside your accounts and you retain full ownership of your data and campaigns. Your data is used only to operate your campaigns and is never sold. ## Measuring and improving AI visibility **How does Lily prove the results are real?** Lily measures every change against a control. Using A/B and difference-in-differences testing, Lily isolates the lift caused by better product data, so results reflect the enrichment, not seasonality or spend. Tested, not promised. **Can Lily guarantee my products will appear in ChatGPT, AI Overviews, or other AI assistants?** No. No tool can guarantee placement, ranking, or citation on an AI surface, and Lily does not claim to. Lily Max improves the readiness and quality of the product data those systems read, which is the part you control. **Can I measure revenue from AI-driven discovery?** Yes. Lily tracks how AI-driven discovery and agentic surfaces contribute to revenue, so you can see the impact of product intelligence beyond traditional channels. **Why does Lily say "the score is not the outcome"?** A visibility score is a means, not an end. Lily Max uses it to find the product signals most likely to move visibility, conversion, and ROAS, then decides what to deploy based on measured performance. ## Customer outcomes and feedback **What kind of results do customers see?** Outcomes vary by catalog and starting point, but customers typically see meaningful lifts in agentic and paid-channel revenue without proportional increases in ad spend, because the catalog itself does more of the work. **How long until I see impact?** Most teams connect their data and launch within days, and begin seeing optimization impact as Lily Max iterates on product intelligence across channels. ## Implementation, security, and governance **Does Lily Max replace my existing tools or PIM?** No. Lily Max works with your existing feed, PIM, commerce platform, and agency. There is no replatforming. It improves the quality of the product data those systems already distribute, so the tools you run perform better. **How do I get started, and how soon will I see results?** Getting started begins with connecting your product feed and scoring your catalog to find the highest-impact gaps. Most teams launch within days and see validated results as the first controlled test resolves. Timelines vary by catalog size and data quality. **How long does onboarding take?** Most teams connect their catalog, ad accounts, and storefront and launch within days. We work alongside your team to scope a pilot and roll out across channels. **How does Lily protect my data?** Data is encrypted in transit and at rest, access is least-privilege, secrets stay server-side, inputs are validated, and access is logged. See our Security & Data Trust page for more. **Is my catalog or ad data ever shared or sold?** No. Your data is used only to operate your campaigns and improve your results. It is never sold. ## For agencies **Does Lily work for agencies managing multiple brands?** Yes. Agencies use Lily to run catalog-aware retail media and product intelligence across multiple clients, with each brand's data kept separate and campaigns operated inside the client's own accounts. **How do I partner with Lily as an agency?** Reach out through our demo form and tell us about your clients and channels. We'll walk you through how partnering with Lily works.