CEO Corner: Purva Gupta Wants to Bridge the Gap Between Consumer and Merchant Speak
When Purva Gupta emigrated to the United States from India and did some online shopping here, she found herself surprisingly frustrated. Why was it so hard to find the specific type of dress she was seeking? Thinking it might be an immigrant problem resulting from a language or cultural barrier, she did some hefty local research to “test her hypothesis.” But after speaking with more than a thousand women, she confirmed there was definitely a disconnect between how consumers experienced and perceived products, and how brands and retailers described such products. This compelled her to create an AI-driven solution that bridges the merchant-consumer gap.
Sourcing Journal caught up with Gupta to discuss Lily AI, which she co-founded with chief technology officer Sowmiya Chocka Narayanan. Lily AI recently received a $25 million Series B investment, which it is using to expand into mid-market retail e-commerce brands across home, beauty and fashion.
More from Sourcing Journal
Virtual Try-on Startup Zelig Scores $15M, Eyes Personalized Marketing
SJ Fall Summit: AG CEO Glenn McMahon Plots International, Domestic Expansion
Sourcing Journal: How did your prior background lead you to co-founding Lily AI?
Purva Gupta: While I’m an economist by education, my experiences working in India at Saatchi & Saatchi and then at a startup paved the way for what became Lily AI. While at Saatchi & Saatchi, I was inspired by the power of the emotional connection between a shopper and a brand. I realized I wanted to work in technology and change people’s lives, and concluded that if I wanted to create my own company, then the core of the problem it was trying to solve had to be a problem I experienced personally and deeply connected with.
When it comes to online search engines, how big is the gap between merchant speak and consumer speak?
From a language perspective, the words real people use are far more colloquial and often more nuanced than the standard words used by retailers and brands to describe their products. Consumers have unique emotional contexts and perspectives. When they detail what they’re looking for, they use a rich, personalized vocabulary that includes dimensions like trends, occasions and styles. Ultimately, it boils down to product details, which, in merchant-speak, specifically refers to the product attributes that exist in a retailer’s product taxonomy.
For example, a brand may tag what a consumer calls a “summer wedge” as a “supple leather upper resort wedge sandal,” a “back-supporting mattress” as a “perfect sleeper ultra-plush hybrid gel mattress,” or a “lightweight summer foundation” as “Stay-in-Place Flawless Wear Cashmere Matte Foundation.” The examples are endless but show how consumers and retailers approach language differently.
How do you train Lily AI’s algorithms to be more in step with how consumers search?
Humans are always in the loop! Our Domain Experts have backgrounds in retail (merchandisers, stylists, marketers), and this team stays on top of the latest trends to inform the most robust consumer-friendly product taxonomy resulting in improved trend discovery. They are constantly conducting in-depth research into micro- and macro trends, textile and color trends, and social media trends. Armed with this information, they train machine learning to ensure the Lily AI product taxonomy is built to match consumer trends with relevant attributes.
Data scientists and AI engineering experts are also a part of the many humans behind Lily AI’s unique “consumer-oriented” product taxonomies vs. pure-play automation, constantly refining models and ensuring the absolute best in data quality, accuracy and relevance. This combination of experts is constantly training and refining the algorithms, resulting in an ML that matures over time, continuously getting ‘smarter’ and ever more accurate with every training input.
What are the results of this?
We have compiled a proprietary library of over 20,000 consumer-oriented words spanning attributes, synonyms and trends, and we use this ever-expanding, vast data asset to inform our product taxonomy. Doing so, we can keep pace with the evolving voice of the consumer. Amongst some of our brands, which we are not at liberty to disclose, we have seen a 3.5 to 9 percent increase in online order conversion, a 2 to 5 percent increase in product detail page (PDP) views and a 3 to 10 percent increase in demand.
With GenAI in general, users are learning the value of an expert prompt. Do you think this growing expertise will help improve online shopping searches?
One shouldn’t need to be a prompt engineer to find what they want. The great news here is that search engine technologies and platforms will continue to evolve so that consumers don’t need engineering degrees to shop online! We are in the early innings of GenerativeAI, and as we have already seen in the one year since ChatGPT launched and changed our world, it will only get better, not to mention, safer.
But even with great prompts, for search to “find” what a person seeks, we still need the relevant product details and attributes to be properly labeled to power the discovery.
How does Lily AI help with demand forecasting and what have been some tangible client benefits?
Planning and forecasting are prioritized focus areas for many of our clients due to the massive margin increases to be realized from improved pre-season and in-season models. At Lily AI, our demand attributes help retailers to enhance product design, improve replenishment and allocation models and deliver an assortment that maximizes margin opportunity.
One of our multi-brand clients projected $7 million to $48 million in topline revenue increase from leveraging the Lily AI-improved product attribution data in their forecasting models. Another global retailer estimated a potential to reduce weighted average percentage error (WAPE) by ~20 percent and improve gross margin by $300 million across all brands.
How is AI evolving and how can brands and retailers harness it to maximum effect?
AI for retail is not new. Be it data-driven analytics, applying machine learning in inventory planning, supply chain all the way to powering customer experiences through recommendations, chatbots, and detecting anomalies in retail security, machine learning has played a role in retail for quite some time.
The underlying technology has been evolving rapidly and getting smarter by the day, and the wave of deep learning excites us for its ability to learn to make connections between input and output and requires less spoon-feeding than earlier ML techniques needed. And now Generative AI has pushed ML capabilities from analyzing or classifying existing data to being able to create something entirely new, including text, images, audio, synthetic data, and more.
That said, in order to effectively harness the value of today’s powerful suite of AI, it is important to always start from deeply understanding the use case and the problem we are trying to solve, having the right, accurate data, and then the skillsets of the team and infrastructure to be able to experiment to arrive at the right solution.
At Lily AI, we perform thousands of experiments before we push the outperforming models into production. Our platform is also built with the flexibility to swap in/out the right models for the problem in hand. Our vision is to bring humanity to shopping and we are excited to continue to innovate and draw on our Retail AI expertise to help global brands and retailers thrive.
