Joyce Lay is a veteran Lead Merchandise Analyst at Lily AI, and she’s been instrumental in helping to build product taxonomies and in creating the universal mathematical language that powers the Lily AI customer intent platform. She’s now turned her attention to using her expertise to do the same for beauty brands and retailers. We caught up with her to understand how beauty products differ from fashion/apparel products, the challenges faced by beauty brands in converting intent into purchases, and more.
You’ve been working as a styling expert at Lily AI on fashion and apparel for a long time now. What sort of key differences do you see in the beauty industry when it comes to product data, and how consumers search for (and find, or NOT find) what they’re looking for?
Joyce Lay: Like apparel, shopping for beauty products is also a personal experience for a consumer. Whether it’s because the product fits their lifestyle or because it has a wide variety of skin tone colors, beauty shoppers search for products that are a reflection of themselves. However, I’ve noticed that it’s usually driven by the consumer trying to tackle what they want to prevent, remove, or hide - things that they don’t want visible, such as wrinkles, dark spots, stretch marks and gray hairs.
Makeup is similar to apparel because it showcases a person’s personal style, or is informed by current trends. But developing that personal style usually takes quite a bit of practice or trial and error that not everyone is open to. I’ve also found that beauty consumers are extremely loyal to their preferred brands or products. If they like it and it works for them, they’ll stick to it.
What have our retail customers told us their biggest challenges are when it comes to the beauty products in their ecommerce stores? What are they looking to solve for?
One of the biggest challenges when shopping for beauty products online is not being able to use our sense of touch and smell. This is even happening today with COVID restricting in-store testers. Consumers want to know how the product will feel on their skin not only on its own, but also when layered or mixed with other products. They want to know if using this formulation type or ingredient mixed with another will work well or not. This is something that’s very difficult to portray through text, because no brand wants to share what doesn’t work well.
At LIly AI, we’re currently able to capture touch through the types of formulation, and sometimes even smell by the ingredients or notes (for fragrance) that are listed. Examples of ongoing projects the team is working to better understand these sensory challenges are:
understanding what mixes of formulation types will create what type of feel, or how it will sit upon the skin
identifying which notes listed in fragrances link to the type of person or overall scent the consumer wants to define themselves as (romantic, energetic, sophisticated)
Another challenge is finding the right skin-tone color when purchasing complexion products. Brands have advanced today by offering a wide range of colors and undertones, but it opens up new challenges. Which one is my skin color? Is my skin undertone neutral or golden? Will this foundation oxidize as I wear it throughout the day? Fenty’s Foundation is a great example of a complexion product that provides a lot of color and undertone options, yet for those who don’t know where to start when purchasing foundation, this can be overwhelming.
What are the core beauty products that Lily AI has built a taxonomy for?
Lily AI currently has extensive taxonomies built for skincare, bath and body, face makeup, eye makeup, lip makeup, haircare, hair tools, makeup brushes and tools, spa tools, nails, and fragrance.
What are some of the challenges you’ve found in building taxonomies for, say, “clean beauty” products, or for other products?
Clean beauty is still a challenge we’re facing. We found that with the rise of popularity for clean beauty, many retailers/brands have just done the bare minimum in defining what it means. What we’re trying to do by understanding ingredients is to differentiate between all the different types of clean beauty out there, such as vegan beauty, cruelty-free and sensitive skin-friendly, rather than grouping them all in one generic category.
Even though it’s great that there are usually many photo options for a makeup product, sometimes its color can look extremely different in each image. For example, if the lipstick is worn on the model’s lips, it can look quite different when it’s in the packaging. That’s because we already have a natural tint on our lips, so when lipstick is applied, it can look different on everyone. As we've discovered, makeup swatches or product images (without a model) most closely resemble the truest form of color. Here are some examples:
Not determinable images:
Tell us something surprising you’ve found as you and your team have defined our beauty product attribution data.
We definitely expected beauty products to be more text-driven, but we were surprised at just how many alternative ways an attribute can be named. For example, lip gloss can also be known as lip glaze, lip shine, lip topper, liquid gloss, and tinted gloss. Based on the shopper's intent, Lily's beauty taxonomy will be able to provide synonyms that will enhance the retailer’s product recommendations.
What do you see on the horizon in this industry, and how will Lily AI help our beauty ecommerce customers address them?
Brands have improved by being more transparent with their ingredient lists, and by providing more inclusive and relatable images to reach out to its audience (different skin tone models, including older models, and even influencers). I predict they will continue this route to cater to their audience. As Lily AI focuses on capturing all the details in each product, I think we can help provide more accurate product recommendations not just on the beauty brand’s ecommerce site, but for their customers as well. We will be able to tap into each need, skin-type, or desired result specifically as if it’s a 1-on-1, in-person experience with an expert, rather than a general recommendation.
Would you like to talk with a Lily AI specialist about how your brand can dramatically improve site search, personalized product discovery, recommendations and demand prediction? Let's talk!
Understanding Products and Customers with AI: A Q&A with Lily AI’s Matthew Nokleby
What are some of the latest breakthrough advances in AI, as well as current data science challenges in retail? We asked Matthew Nokleby, Lily AI’s Tech Lead Manager, Computer Vision, for his expertise on these questions, on building ML models, and advice he’d give right now to those pursuing careers in data science.
3 Reasons Why Ecommerce Product Recommendations and Personalization Urgency Has Never Been More Pressing
Two years into the pandemic, we’ve certainly learned a few things about why ecommerce success is so tied to having strong product recommendations, online personalization and rock-solid demand forecasting. Here are 3 core reasons why ecommerce product recommendations and personalization urgency has never been more pressing than it is today.