Walmart’s Dr. Astha Purohit: getting innovation right has quietly become table stakes for retailers
RTIH talks AI trends, challenges and opportunities with Dr. Astha Purohit, Director of Product Management, Customer Data and Identity at Walmart, and a member of the RTIH Retail Technology Hot 100 List judging panel.
RTIH: Tell us about yourself.
AP: I was born and raised in India. I am a physician by training and got my MBA from MIT. Post-MIT I joined McKinsey focused on the technology practice for the retail sector.
I've spent the last decade working in retail tech across companies like DoorDash, CVS, and Walmart. I've been predominantly focused on the data and infrastructure layer - the foundational ecosystem that powers online shopping and lets customers search, compare, and check out.
Outside work, I'm a voracious reader, enjoy going on long walks, and love travelling.
“I think a lot of people are feeling a real sense of AI fatigue.”
RTIH: For the last 20 years, retail technology has been built around a person browsing a website or walking a store. That's quietly starting to change, as AI agents start acting on behalf of the shopper. Could you talk about this shift?
AP: As I think about AI agents acting on behalf of the shopper, the important thing is to see how the retail experience shifts depending on the segment.
When it comes to luxury purchases, where a customer is spending a lot of money, increasingly it's going to be about experience. I think we'll see a shift towards in-person, analogue experiences here. It becomes even more experiential and shopper centric, and brands will lean hard into that.
AI will largely not be part of this retail segment, especially from the customer's point of view. The tech will leverage AI, but the customer experience will stay personal, and brands will focus on brand building and experiential shopping experiences.
Next is the considered purchase segment - a laptop, a television, an expensive appliance. Here we'll see heavy AI use: comparison research, pulling reviews, narrowing the options. But the final decision will be made by the customer. It will depend on whether they saw an influencer they follow using it, whether they trust the brand, how they feel about spending the money. The final decision-maker is the customer.
Then there's everyday shopping. Most of this gets automated over time. Agents build the basket, assemble the grocery list, push it to your app so you can add things in a tap, reorder the staples. It moves towards a subscription rhythm - you're barely thinking about it.
So as agents start acting for shoppers, it becomes increasingly important for retailers and brands to understand which segment they sit in - premium, considered, or everyday - and design the shopping experience around that. The right approach in one tier is the wrong approach in another.
RTIH: What are the key challenges facing retailers when navigating this fast changing landscape?
AP: I see two big challenges. The first is catalogue debt. Most retailers carry years of product data that was good enough for a human navigating the page - inconsistent taxonomy, missing attributes, at times, variants that aren't on the same page, missing spec or nutritional fields you'd have to Google. Humans could interpret the page.
With agents entering the fray, clean data has become important - but cleaning up a large catalogue, while genuinely painful, can be done with AI. You can use models to normalise taxonomy, collapse duplicate products, enrich attributes, and find and fill in nutritional profiles at scale.
The second is discoverability - what's becoming known as generative engine optimisation (GEO). For 20 years the game was ranking on page one of Google. The new question is different: what makes an AI agent pick you? And here's the part retailers underestimate.
When someone asks an assistant for, say, a good vitamin brand, the model isn't just reading your website and making a decision - it's synthesising what the wider web says about you. Structured product data gets you parsable; but what gets you recommended is consistent, credible third-party signal - reviews, expert roundups, real discussion in the places people talk, like Reddit. An agent treats that community sentiment as a trust proxy, the same way a person asking a knowledgeable friend would.
So, the two challenges meet in one place: clean structured data makes you legible to the agent, and authentic outside signal makes the agent choose you. Neither alone is enough, and there's no shortcut around either.
“The retailers and brands that keep a genuinely human layer hold something an agent can’t easily replicate, and I think that becomes a real source of durability.”
RTIH: What are the key benefits for retailers who get this right?
AP: I think getting this right has quietly become table stakes. Over the next couple of years, as agents mediate more of how people shop, GEO becomes the baseline everyone in retail has to meet.
For 20 years, retailers competed for the customer's attention - the search result, the shelf, the homepage. Increasingly, you're also competing for the agent's attention, because the agent decides which handful of products the customer ever sees. Getting this right is going to be critical for retailers to win in the coming decade.
RTIH: What in your view will be the AI related trends and developments to watch over the coming months and into 2027?
My top three are as follows.
First, the fight over who owns the customer. As more shoppers start their journey inside third-party AI chatbots, the chatbot - not the retailer - becomes the front door. That's a real risk for retailers: you can quietly slip from being the destination a customer chose to being an anonymous supplier the agent picks from. I think the defining strategic question over the next couple of years is how retailers keep a direct relationship with the customer, through loyalty, through exclusive products, through reasons a shopper comes to you vs letting an agent decide for them.
Second, discoverability becomes a real discipline, measured and tracked. Right now, "how do I get picked by an agent" is something most teams are improvising on. I think over the next two years it hardens into an actual practice, with its own tooling, its own metrics, its own methodology - the way SEO did over the past two decades. And it becomes something you can actually measure: not just "are we ranking," but "how often does an agent surface us, in which categories, and against whom."
Third, with so much AI all around us, I think a lot of people are feeling a real sense of AI fatigue, and it swings value back towards human trust and in-person experiences. A recommendation from someone you know and trust, a friend, a creator who isn't perpetually running ads, starts to mean more, precisely because it's human.
So, the smart move isn't only to optimise for AI agents. The retailers and brands that keep a genuinely human layer - real voices, real relationships - hold something an agent can't easily replicate, and I think that becomes a real source of durability.
