Thursday, August 18, 2022
HomeDigital IconsInterview with Purva Gupta from Lily AI

Interview with Purva Gupta from Lily AI

Team eCommerce Next interviewed Purva Gupta from Lily AI to get more insights on what retails should know about the future of product taxonomy. Following is our interview with him:

What is it that AI brings to the table when it comes to product recommendations for the ecommerce consumer?

We work to solve the all-too-common problem faced by consumers when they’re shopping on ecommerce sites: finding what they’re actually looking for. AI, deployed intelligently, can show a shopper what they want online before even they know they want it.

Our AI turns qualitative product attributes like fit, fabric, occasion and style into a universal mathematical language. By using AI-powered image recognition, Lily allows retailers to improve their on-site search, personalized product discovery, recommendations and demand prediction, the combination of which translates into 8-9 digits lifts for the retailers who use our platform.

How have Bloomingdale’s and others used Lily AI to achieve their ecommerce goals?

Bloomingdale’s were looking to improve product discovery, and reasoned (correctly) that by automating the product attribution process, they would increase the relevance of searches for each individual shopper. They leveraged Lily AI’s consumer-focused product attribution tags to augment their on-site search across apparel, shoes, handbags and jewelry, and very quickly increased the relevant results for descriptive searches by up to 30x. When a consumer is able to find what they’re actually looking for, even in a long-tail search like “boho chic party dress”, it breeds higher loyalty, better on-site conversion and a boost in revenues.

Many are using AI to create product recommendations, so what’s it take for companies just like yours to break the barrier for mass adoption?

Frankly, you need to have a unique angle, because the space is saturated with recommendation products that deliver varying levels of results. It starts with depth of knowledge of product data. Other recommendation engines rely on data that is input from the retailers themselves – sometimes this is as simple as color, size and a few other fairly generic inputs. We’ve found time and time again that this is typically very thin and inconsistent, so therefore recommendations that are built off of that data will be incomplete, and result in some of the cognitive dissonance you see on ecommerce sites, where you’re recommended something that has nothing to do the item you’re currently shopping or your own tastes and style.

Great AI requires great data to be used as an input from the start . Lily’s ability to enrich products with 10-20 attributes per product enables us to provide a much richer set of inputs into the AI model, and this then produces a far more relevant set of product recommendations to the consumer.

Where will product recommendations be in three years? What does that look like?

The entire shopping experience will be different in 3 years. AI and computer vision helps retailers understand intent and deliver better recommendations today. In the near future, we’re likely to be more connected with shoppers, and there will be a more of a free flow of information and a sharing of their preferences that can be leveraged intelligently across channels to deliver more personalized experiences and build better retailer/consumer relationships.

There are many marketing promises about this vision today, but most products don’t make it easy to deliver on that promise to a point where it actually makes a difference in the day-to-day online shopping experiences of the consumer.

What does the consumer gain from adoption of this by brands?

Shopping is a visual activity. We shop based on what we see, not necessarily on what we type. Consumers, whether we know it or not, are intuitively looking for focused, relevant suggestions that evolve our previous search history and resonate with our current contexts.

It’s no longer acceptable for a retailer to return 10-plus pages of search results that drop off dramatically in relevance after the first couple of results. By using this granular product data and enabling long-tail semantic searches and predictive autocomplete, brands can actually guide shoppers directly to the items they really want.

Which brands are willing to bet /take a chance on AI for product recommendations?

I suppose a related question might be why a retail brand would choose to still rely on simple tools like text-based keyword search and spreadsheet-based demand forecasting when consumers have clearly been trained by Netflix, Spotify, Amazon and others to demand a highly relevant, highly personalized “shopping” or entertainment consumption experience – which very much includes recommendations.

We’re very focused, along with our retailer partners, on the marriage of product and consumer intelligence, which matches our deep tagging of products (with over 15,000+ attributes) with deep profiling of consumer motivations. Our AI doesn’t just use this to power recommendations, but to send this universal language of attributes to all destination systems in the retail technology stack, helping to power search engines, demand prediction models, item set-up processes and more.

Regarding attributes, there’s little context for what that means to consumers. What’s the cost of less than 10x attributes and why does this matter?

There’s a true cost in missing the “long tail” of searches and in only being relevant for the most basic search, i.e. “white crew-neck t-shirt”. One person might see a dress and think “New Year’s Eve”. Another sees the same dress and thinks “metallic strapless dress”. If you don’t have those sorts of attributes in your product taxonomy, you won’t be able to serve up the right result to either person. Retailers deliver much more relevant results by having more attributes.

Regarding consumer privacy issues: do consumers know about the mining of attributes and do they care? Or is this all in the course of doing business today?

It’s actually much better for shoppers because it is product data, not personal data. We infer people’s preferences based on product attributes and anonymized behavioral data, and not by trying to pry into their personal data. I think people are at the point of expecting brands to know and remember what products they’ve browsed and bought these days. Think of Amazon and “Buy It Again”. In fact, it’s when brands do a poor job of piecing your prior experience together that it leads to poor shopping experiences and a loss of loyalty.

About Purva Gupta

Purva Gupta is the co-founder & CEO of Lily AI, the first customer intent platform built to power the present and future of ecommerce. Lily AI injects robust product attribute data and unique customer intent into the entire ecommerce stack, supercharging retailers of all types by dramatically improving on-site search, personalized product discovery, recommendations and demand prediction, unlocking millions in new revenues.

She previously worked at UNICEF Ventures/Innovation Fund investing in lifesaving apps & technologies, and also led marketing efforts for Eko, a branchless banking & mobile payments startup. Purva is a Tory Burch Fellow 2019; an immigrant founder who has been on 6 visas in the last 4 years to live her American dream, and who has an MBA from Indian School of Business and a Bachelor’s in Economics from Shri Ram College of Commerce.

About Lily AI

Lily AI is the customer intent platform built to power the present and future of ecommerce. We inject robust product attribute data and unique customer intent into the entire ecommerce stack, supercharging retailers of all types by dramatically improving on-site search, personalized product discovery, recommendations and demand prediction, unlocking millions in new revenues.

RELATED ARTICLES

Most Popular