The Center for Data Innovation spoke with Purva Gupta, co-founder and CEO of Lily AI. Lily AI uses an advanced algorithm to refine product attribution, improving product search and recommendations on e-commerce websites. Gupta explains how AI can improve retailer decision-making and reduce bottlenecks in the product supply chain.
Becca Trate: What issues did you identify that led to the founding of Lily AI?
Purva Gupta: It really all started with personal frustration when I arrived in the United States from India at how fashion retail was failing to understand the amount of detail that shoppers actually use in real life to describe the items of clothing, accessories, or shoes they’re looking to buy. It felt to me that the clothes themselves were at the center of the entire experience, and it wasn’t really the shopper; instead, it felt like brands were saying, “I have this item of clothing, come into the store; if you can fit into it, great.”
My co-founder, Sowmiya Choka Narayanan, and I founded Lily AI in 2015 to build a shopping experience that would be able to understand the emotional context of the shopper in a way that had never happened in online commerce or in retail before. We have extended this concept over the years into an enterprise-grade, AI-powered product attributes platform that injects the language of the customer across a retailer’s existing stack. We’re driving eight- to nine-figure revenue uplift for retailers and brands like The Gap, Bloomingdale’s, Macy’s, and thredUP by dramatically improving their on-site search conversion, personalized product discovery, and demand forecasting, among other retail applications, and with more coming shortly.
Trate: Can you explain product intelligence and how it improves online product searching?
Gupta: It starts with understanding the language of the customer, and realizing just how poorly served they are in the online shopping experience today. Products are often put on shelves with legacy, “out-of-the-box” attributes that come directly from manufacturers and distributors, and that don’t capture the nuance and detail that shoppers actually use when they’re looking for relevant products that match their needs.
Different shoppers search uniquely, so that really makes it essential for retail e-commerce brands to build the right product taxonomy to capture both common and long-tail searches. Think of your own frustrating experiences on retail e-commerce sites and receiving irrelevant results or worse, no results at all, even when the product you’re looking for is clearly carried by that retailer. What do you do? You move on to a different shopping site to find what you want, and there’s a good chance you don’t return to the original retailer because they’ve already proven they didn’t have what you wanted—even if they actually did, but couldn’t surface it because they didn’t inject that language of the customer across their retail stack.
Trate: Lily AI offers solutions for “demand forecasting.” What role does AI play in forecasting?
Gupta: Supply chain problems have exposed retailers, so they need to make decisions on what merchandise to order even earlier than they normally would, in order to keep shelves—both digital and physical—stocked.
Having better and more granular product attribution data is the key to overcoming bottlenecks and problems. It ensures the right size, color, and style mix of items will still be ordered ahead of much longer lead times. Having deep product data allows retailers to order earlier, make predictions earlier, and ensure that the decisions that they make are more precise—helping to overcome the fact that they now have to make those decisions much earlier than they’d like or have previously done.
Without it, retailers run the risk of not being able to sell what they order at full margins and will need to sell through “bad guesses” at a discount. Gross margins will suffer. Worse, they can underbuy, and lose the ability to sell products that would otherwise have sold at full margins.
Lily AI provides the best possible data with the best granularity and highest accuracy. Better comparisons, better relationships between products, and higher product attributes lead to better ordering.
One multi-brand retailer with whom we work was able to reduce its forecasting timelines from three months to one month with automation at scale. Lily AI was able to replicate the proxy product predictions of their best merchants, now relying on data that drove even higher accuracy levels. This increase in accuracy and time savings in the forecasting pipeline resulted in the right products being ordered at the right time, as well as the ability to get ahead of supply chain orders and sell more products at full margin. This is projected to positively impact this retailer’s topline revenue this year by up to $48 million.
Trate: Consumers are balancing both e-commerce and in-store shopping. How is AI used to transform the in-store experience?
Gupta: This is where AI-powered demand forecasting is particularly relevant. Knowing why groups of shoppers bought something in the past—for instance, because that “floral print black dress with lace” had lace on it, not because it was black—gives retailers some very powerful signals about what to stock shelves with, both physical and digital.
Gathering proxy products through Lily AI-powered computer vision helps retailers to accurately forecast demand for brand-new product lines. We work to help them replace wholesale pre-orders with a leaner, demand-led, made-to-order model that fuels product development with AI to launch lines that can be sold at full margin, rather than discounted later.
Trate: What do you think the future of AI in retail looks like?
Gupta: I think there are many incredible applications for AI in retail and it’s exciting to see them optimizing both the in-store and the online shopping experience, as they roll into our daily retail interactions without us even knowing it. Yet at the end of the day, there’s still that fundamental disconnect between retailers and shoppers that we saw even back when we founded Lily AI: the need to have a core layer of customer language—that expanded taxonomy of product attributes—that accurately connects a retailer’s shoppers with the relevant products they’re looking to buy.
Without this fundamental input across the stack, retailers and brands are still making bad guesses about products and inventory, and can’t meaningfully break through the average 2.5 percent conversion rate from online search.