How Agentic Commerce Is Reshaping Discovery, Impulse, and Personalization for Brands

E-commerce today feels like a very product-search-heavy experience, as Kyle McWhirter of Walmart Connect put it at CES 2026. A customer knows what they want, types it in, and buys it. McWhirter opened a conversation about what changes when AI allows discovery to begin with an occasion, a need state, or a habit a customer wants to build instead of a product name.

McWhirter was joined by Shakir Moin, President of Marketing at Coca-Cola, and Amy Andrews, President of Mars Commerce (Mars United Commerce, now part of Publicis Groupe). The three spent most of the conversation exploring what agentic commerce means in practice for one of the largest CPG companies in the world, for the agencies that support brands on the platform, and for the retailers they partner with.

This article draws on that CES 2026 Content Studio conversation between McWhirter, Moin, and Andrews, supplemented by industry data where noted.

Discovery is no longer just product search

Andrews offered the most concrete example of what AI-driven discovery looks like right now. She talked about asking ChatGPT that morning what she should do for her skin in Las Vegas because of the dry air. The response went beyond skin care. It told her to get more sleep, drink more water, and then, because it knew she had purchased Smart Water before, offered to have it delivered to her hotel room.

That kind of experience, Andrews said, represents a contraction of the funnel and discoverability based on all of her past purchases and past prompts. The system moved from a personal need to a product recommendation to a purchase opportunity in a single conversation.

Moin approached the same shift from Coca-Cola’s vantage point. As a global company, he said, Coca-Cola has watched e-commerce evolve into social commerce, particularly in China, where creators have used e-commerce as a medium to build brand equity. It is not only a search engine for looking for a product and buying it at the best price, Moin said. It is becoming an equity driver as well. McWhirter summarized the convergence: performance, brand building, and sales driving activities converging together.

McWhirter connected both perspectives to Walmart’s business. Personalization is playing a key role in making AI more effective at solving everyday problems for consumers, he said. At Walmart, that translates to using recipes and meal occasions to grow the total retail footprint, with beverage as a major opportunity.

Why this matters for suppliers: Andrews’ example is not hypothetical. ChatGPT processes roughly 50 million shopping-related queries per day, according to an OpenAI and Harvard working paper reported by Modern Retail. That is 50 million daily interactions where a customer describes a need and an AI recommends products. If your brand does not appear in those conversations, the customer never sees your product, because there is no shelf to browse and no search results page to scroll. The question for suppliers is whether the information available about your products, on your product detail pages, in reviews, in third-party content, is rich enough for an AI agent to recommend you when a customer describes a problem your product solves.

From mass reach to de-averaging at scale

Moin gave the most detailed account of how a major CPG company is restructuring its marketing around this shift. For the past 70 years since the explosion of television, consumer product companies like Coca-Cola have excelled at scale. Big ideas, great ideas, ensuring the maximum number of people see them again and again.

What Coca-Cola is moving toward is what Moin called de-averaging at scale. The company’s advantage remains its reach across the country, but the ability to use that reach to be more personalized is the shift. Speaking to every one of 330 million Americans individually is not yet feasible, but having a single conversation for all 330 million is not effective anymore either.

Coca-Cola has evolved to looking at the country at a zip code level, understanding consumer needs in specific locations across a portfolio that includes Coca-Cola, Fairlife milk, Smart Water, and Body Armor. The goal is to identify the right solution for a specific occasion and connect that message to commerce in a way that creates value for both the brand and retail partners like Walmart.

Moin went further. Human behavior across a week means what people eat over a weekend is different from Monday versus Tuesday versus Wednesday, and consumption patterns differ across socioeconomic groups. The idea of recipes making meal recommendations relevant by day of the week is something Coca-Cola and Walmart have been discussing together.

From the agency side, Andrews offered a different framing. For all of our clients, she said, we have always been looking at how to win across every single shelf. The agentic shelf is really no different. But the language is changing. In national marketing, brands focus on the reason to believe. In the shopper space, it has been the reason to buy. Now, Andrews said, the question is the reason to belong in the conversation. Not just one conversation, she said, but all of the conversations happening across individual shoppers.

Why this matters for suppliers: Moin’s “de-averaging at scale” concept is not just a Coca-Cola strategy. It points to where retail media is heading for any supplier with a portfolio of products that serve different occasions in different markets. The zip-code-level targeting Moin talked about requires product and occasion data that most suppliers do not currently organize at that level of specificity. If your media plan treats the entire U.S. as a single audience, or even as a handful of regional segments, the platform’s ability to personalize around your products is limited by the data you give it. Andrews’ “reason to belong” framing is the other side of the same coin: your brand needs to be present in the AI-mediated conversations where purchase decisions are forming, and that presence depends on the depth and specificity of your product content.

What impulse looks like in an agentic world

McWhirter raised the question of impulse shopping directly. Impulse is a big part of Walmart’s brick-and-mortar business, he said, but less prevalent in e-commerce. He asked whether agentic commerce creates a new version of it.

Moin broke the question into three distinct purchase behaviors. The first is deliberate and functional, where a consumer researches ingredients, compares products, and makes a considered choice. He used Core Power protein as an example. The second is what Andrews’ Smart Water experience illustrated: a solution to a problem that the consumer did not set out to solve. I do not know how to even categorize that whether it was impulse or plan, Moin said, because it was neither. The third is what Moin called true impulse, where a shopper knows exactly what product they want and hopes it is available at the right price in the right form.

All three behaviors will continue to exist, Moin said. But brands need to be clear on the attribution model behind each one and understand the key drivers by product category and product type.

Andrews pushed the idea further. She pointed out that the conversation was still defaulting to shoppers prompting AI, and described what a true agentic impulse purchase could look like. If the system knows a shopper would be interested in a new Core Power flavor, and that shopper has opted into an agent shopping for them, the agent could buy it and send it to them at a certain price point without the shopper hitting a buy button. Smart Water could be waiting in her next hotel room without her placing an order.

Moin reinforced that with a personal story. More than a decade ago, he checked into a hotel in Singapore after moving from Malaysia. He loved M&M’s then and still does, he said, though he is watching his sugar now. The hotel had placed three packets of M&M’s and two Diet Cokes in his room with a welcome letter from the general manager. The hotel chain in Malaysia had shared his preferences with the Singapore property, and that experience changed his perception of the entire chain. For brands to do something similar at scale, Moin said, across an entire consumer base rather than one hotel chain, is what this new technology makes possible.

McWhirter connected the thread: the underlying data is what informs AI and the opportunities to personalize an experience. Without good data, he said, you are limited in what you can achieve.

Why this matters for suppliers: Andrews’ distinction between a shopper prompting an AI and an AI agent purchasing on the shopper’s behalf is the most important strategic distinction in this conversation. The first scenario is a new version of search. The second is a new form of distribution. If a customer has opted into an agent making purchases at a set price threshold, the brand that the agent selects is the brand that wins, and that selection depends on product data, past purchase history, and how well the brand shows up in the information the agent consults. Suppliers building agentic commerce strategies should think about both scenarios separately, because the content and data requirements for each are different.

Feed the agents more, not less

All three speakers returned to the same point about how brands should operate in this phase. Andrews said the approach is the opposite of what marketers have been trained to do. We have spent so much time as marketers trying to be concise, she said. This is almost the opposite strategy.

Her example was specific. On product detail pages, brands have always tried to find the perfect word or image to drive conversion in three seconds or less. Now there is an opportunity to upload user manuals into the text space. The shopper might not have time to read that content, but the agent does. The agent takes all of that content and can tailor it to the different conversations that consumers are having.

Andrews also pointed to Reddit as part of a commerce strategy, saying that for ChatGPT, 40% of the sources are coming from Reddit today. The actual data is more nuanced. A Semrush analysis of 150,000 AI citations across 5,000 keywords found Reddit accounting for 40.1% of references across all major LLMs combined. For ChatGPT specifically, Profound’s research found Wikipedia is the top source, with Reddit at number two. But the broader point Andrews was making holds: AI models pull heavily from user-generated content platforms, and brands that treat Reddit and similar forums as part of their commerce strategy will have more surface area in AI-generated recommendations than those that do not.

On experimentation, Moin offered a framing worth repeating. In the pre-AI world, if you made a mistake, the chance of making the same mistake again was always there. With AI, if the AI made a mistake or if a human made the mistake, it would be the last mistake that would ever be made, because it will learn and you would always be better off. The tip for marketers, Moin said, is that nobody has all the answers, but brands need to be in a deep learning mode with all of their partners to find solutions to new opportunities and in some cases new problems. Failure is an option, and the learning ability is allowing us to be better with every single failure.

Internally, Coca-Cola has embraced this. The company has been using the term that disruption should be our playground, Moin said. There was a time when disruption was a headwind. Now internally in Coca-Cola, disruption should be a playground, and agentic commerce is a disruption, and we are going in heads forward.

Andrews agreed on the speed. We always say we want to fail fast, she said, but there is risk in not doing that in this space because we will just be left behind.

McWhirter closed by connecting the thread. Not enough experimentation is worse than failure, he said. And once learnings are applied, they are self-reinforcing. You do not make the same mistake twice. That is how you drive incremental improvement over time.

Why this matters for suppliers: Andrews’ PDP advice is the most immediately actionable point in the conversation. Most suppliers optimize product detail pages for human shoppers browsing on a screen. AI agents read differently. They process more text, weigh specification data, pull from review content, and consult third-party sources like Reddit threads. Suppliers should audit their PDPs with two audiences in mind: the human who needs to convert in seconds, and the agent that needs enough information to recommend your product over a competitor’s when answering a natural-language query. The Reddit point extends this further. If the only place your product’s use cases and benefits appear is on your own PDP, you are invisible to a significant portion of AI-driven recommendations. Brands with organic presence on Reddit, review sites, and community forums will show up in more AI conversations.

Winning With Walmart

Winning With Walmart is an independent platform for suppliers, sellers, solution providers, and industry experts.

Articles are developed by staff editors, contributing experts, and trusted partners.

Some are researched and drafted with the assistance of AI tools and are reviewed, fact-checked, and edited by Winning With Walmart editors before publication.

Editorial judgment, sourcing decisions, and final approval rest with the publication's human editors in every case.

We are committed to accuracy and fairness. If you believe this article contains an error, we welcome your feedback.