How Lily Built an Empathy Engine for Shopping
The mobile app uses AI with emotional intelligence for fashion shoppers.
Empathy in artificial intelligence can mean many things. But for Lily's Sowmiya Chocka Narayanan, machine-- powered empathy in fashion retail translates to one mission — helping customers feel better about shopping for apparel.
“We're building an emotional intelligence-powered shopping experience so women can feel confident about their fashion choices,” said the Lily co-founder and chief technical officer during her session at Re-Work's AI Assistant Summit in San Francisco on Friday. “It's about being that friend you want the validation from.”
For Narayanan, Lily's mission started with a simple insight but one she found shocking: “Girls, babies, at the age of two, start looking at the mirror and making perceptions about their own bodies.” According to the executive, women have 13 negative thoughts about their bodies every day. And, she said, shopping has made this worse. So many women see good stores and great clothes, but still nothing that fits her. This is the problem Lily aims to solve.
The company defines emotions as “the difference between your user's perception about their body and the reality,” she said. “The more positive this is, the more confident they feel about themselves. And the more negative, the more insecure they will be.”
With any AI-based solution, the amount of data matters and so does the way it's categorized. Lily, which works through a mobile app, pulls in 100,000 attributes, tagging each and every product. It also learns about each customer.
“The user chats with
Lily for about two minutes,” Narayanan continued. “We try to understand the user from multiple dimensions: What is the reality? What are their preferences? What is their perception of their body? And then Lily comes up with, like, a confidence score for each part of the body.”
Lily also goes deep to understand specifics such as cut, color, fabric, fit, and how those features can flatter certain body types.
For example, Narayanan pointed to the belts of three dresses. “When it's at your waist, it's not good for people who have a bigger belly,” she said. “But it's great for people who want to accentuate their waist and show more curves.” Likewise, the belt above the waist works for thicker mid-sections, camouflaging and making them look slimmer. The lower waistline flatters shorter torsos, visually balancing out the user.
If such details can be quantified and tagged, then they can power an AI system. Not that there aren't challenges. Sometimes descriptions are insufficient, items can be subjectively tagged by the people or retailers describing the items, and cut-off images can make a blouse indistinguishable from a dress. Cross-promotion can be another wrinkle, if the product blurb includes, say, material and color information for another item. The process is one of constant revision and evolution, with every new piece of data being fed back into the system to improve it.
As it improves, so does Lily's styling engine, which makes the actual product recommendations.
Narayanan is inspired by Hollywood-like high-impact makeovers, which depict women overcome with emotions over their transformed selves. But she's using machine-driven empathy to get there.
—