Most ecommerce brands aren't ready for AI to shop on their behalf

We didn’t set out to find a pattern. Every few weeks we sit down with someone outside Hello Retail for our Conversations series: a Shopify partner, a Klaviyo Community Champion, a growth executive, one of our own data scientists. Different companies, different jobs, no shared script.

But across four of the last seven episodes, unprompted, the guests converged on the same warning.

It isn’t “AI is coming for ecommerce.” That part is already true and mostly uninteresting. The real problem is narrower and more uncomfortable: most ecommerce brands’ data isn’t in a shape that lets AI act on their behalf, not the AI a brand builds and not the AI a shopper brings with them. The advisory layer between a product and a buyer is shifting to a machine, and most catalogs, most inboxes, and most content are not ready to be represented by one.

Browse all Hello Retail Conversations episodes →

The shopper’s advisor is no longer a human

Peter Sommer, CEO of Shopify Plus partner Dtails, put it in terms of who a shopper trusts for advice before they buy anything.

“I found myself actually getting the advice where I simply put up five criteria. It came back with a range of five printers that matched those criteria.” (Peter Sommer)

He wasn’t describing a hypothetical. That’s how he shopped for his own printer: asking an LLM to filter a category he had no expertise in, and buying whatever it recommended. Peter’s read is that this isn’t a checkout-layer disruption. It’s an advisory-layer one, and it arrives before the cart does.

“The larger the customers are, the bigger their product assortment is, and the further up they are in their advancement of being a mature e-commerce merchant, the more you need tools like Hello Retail to fit the needs of their customers.” (Peter Sommer)

The size of a catalog is exactly what makes this hard. A brand with twenty products can keep every fact about them in someone’s head. A brand with twenty thousand can’t, and an LLM answering “which printer fits these five criteria” is only as good as the structured product data it’s drawing from.

Watch the full conversation with Peter Sommer →

Most brands aren’t using the data they already have

Anna Sophie Christensen, Head of Email Marketing & Retention at FABO and a three-time Klaviyo Community Champion, located the same problem one layer earlier, before any AI is even involved.

“People send out to all the subscribers at once, not differentiating the message or the content. And I continuously preach that we need to differentiate, because you have your most loyal customers who have purchased from you 20 times, and you have a lead that signed up a week ago.” (Anna Sophie Christensen)

Most brands aren’t failing to personalize because personalization is technically out of reach. They’re failing because the decisions underneath it (which segment, which product, which moment) never get made, so the data that would make an AI layer useful never gets collected in the first place.

“Already then, you are collecting zero-party data that you can use for the overall structure on how you communicate.” (Anna Sophie Christensen)

That’s the gap in miniature: the readiness problem doesn’t start with AI. It starts with brands not building the structure an AI (or a human) would need to act well on their behalf.

Watch the full conversation with Anna Sophie Christensen →

Personalization without knowing who’s shopping

Sarah Miguel Cournane, a data scientist on Hello Retail’s R&D team, argued the readiness problem has a structural fix, but only if a brand’s product data is built for it.

“If we don’t know who you are, we focus on what you engage with. That’s enough to understand intent.” (Sarah Miguel Cournane)

Her point is that identity was never really the constraint. A brand doesn’t need to know who a shopper is to act intelligently on their behalf, as long as its product catalog is mapped well enough to reason about what any given product relates to. Most catalogs aren’t mapped that way. They’re a list of SKUs, not a structure an AI can navigate.

“We’ve gone from using AI to support decisions to actually verifying what AI produces for us.” (Sarah Miguel Cournane)

That verification step is the part brands underestimate. It isn’t enough to point a generative model at a product feed. Someone (or something) has to keep checking that what comes out the other end is actually correct, on-brand, and useful, at a scale no human review process can sustain on its own.

Watch the full conversation with Sarah Miguel Cournane →

What the AI actually rewards

Jennifer Montague, former Senior Director of GoToMarket at Verdane and now VP of Marketing at Carivo, brought the same problem to a different surface: content, not product data.

“UGC is gold for AI discoverability. That’s what people, that’s what the AI wants to see, is other people talking about you. So it kills so many birds with so few stones.” (Jennifer Montague)

Her argument is that AI systems reward the same thing genuine buyers do: specificity, evidence, a real point of view, and punish the same thing buyers already tune out.

“To stand out, you have to be more authentic, more human sounding. And to do that, you need to have a position that you can lean into that makes you stand out from the bland sea.” (Jennifer Montague)

Generic, unpositioned content doesn’t just fail to convert a human reader. It fails to earn a citation from the AI systems that are increasingly standing between a brand and its next customer, which makes the “bland sea” problem a readiness problem too, just measured in words instead of product attributes.

Watch the full conversation with Jennifer Montague →

The common thread

Take the four apart and they’re about different things: an advisory LLM, an email flow, a product catalog, a piece of content. Put them together and they’re the same warning stated four ways.

AI-mediated shopping doesn’t fail because a brand hasn’t “adopted AI” yet. It fails quietly, upstream, because the underlying data (product structure, customer signal, editorial content) was never built well enough for anything, human or machine, to represent that brand accurately. The AI just makes the gap visible faster than a slow decline in organic traffic would have.

That’s the problem worth sitting with before reaching for a tool. The fix, in each of these conversations, was never “add an AI layer.” It was doing the unglamorous structural work first: differentiating an audience, mapping a catalog, writing something specific enough to be worth citing, so that whatever sits on top of it, AI or otherwise, has something real to work from.

More conversations

Browse every Hello Retail Conversations episode →