I shared in a post last week how every year I make my son a from-scratch cake for his birthday. Last year he asked for a donut cake. So I baked four layers of butter cake, made the frosting, and then drove to Krispy Kreme and skewered a box of glazed donuts on top. At the party, the kids went straight for the donuts. Not one of them were interested in the painstakingly-made cake underneath.
That was the metaphor I used to talk about silos in retail media — the flashy part (partnership announcements! New integrations!) props up a foundation nobody appreciates (measurement fundamentals, tie-backs to sales data, self-serve ad infrastructure).
This week I want to talk about the part of the story I left out: the baking. Those four layers of cake, a sore arm from beating the buttercream. The invisible and unappreciated part.
Because that's the shortcut demon in retail media. Everyone wants to address the exciting stuff, which right now is AI everything. Agentic commerce, LLM-driven discovery, autonomous shopping agents. And the foundational data work that makes any of it possible? That's the baking. It's unglamorous, it's slow, and right now almost nobody wants to fund it.
This is the third and final installment of my series with Anne Hallock, VP Americas at Mirakl Ads, on the demons inside retail media — the internal, structural threats she argues are more dangerous in 2026 than anything happening in the market outside.
Part 1, the growth demon: the top-line scorecard that's started to mislead the industry.
Part 2, the silo demon: how retail media ends up funding tech innovation inside a retailer.
Let's jump in.
This series is sponsored by Mirakl Ads.
Ask them how
Here's Anne's tell for spotting the shortcut. When a retailer announces it's "partnered with" some LLM, the next question is the one nobody asks on stage: how?
"How are you partnered with them? What are you actually doing?" she said. The announcement is the donut. It's the part you can gussy up in a press release and share out on LinkedIn. The how is the part that takes eighteen months of work most people would rather not look at.
And she's sympathetic about it, because the effort is sincere — there's a real scramble across the industry to understand what AI means for retail.
What the boring work actually is
So what is the work?
Anne framed it this way. Think of an LLM as your clone — you're delegating your shopping to something with the power of discernment. You tell it: I'm going to Cannes, I need a maxi-length linen dress, good reviews, and it must arrive by Friday. And the clone goes off to find products that match.
But it can only match on attributes that exist. If nobody on the data side ever tagged dress length, fabric, and made the processing and shipping timeline finable — those products simply don't surface.
And that's the work. Catalogs, taxonomies, attributes, pricing accuracy, review hygiene. The deeply unsexy stuff that determines whether you exist inside an AI shopping trip at all.
And she says when she brings this reality up in a room full of people who came to hear about agentic commerce: "you just see the light die from people's eyes." Nobody dreams of growing up to fix gaps in catalog data.
The numbers back her up. In Flywheel's March 2026 survey of around 40 CPGs, just 31% said they were testing whether their product data can even be read by AI shopping agents — the tagging and formatting that decides if a product shows up in an agent's results at all. Which means roughly two-thirds aren't.
The rules of retail media are being rewritten. Discovery is shifting. The economics are following. And the retailers and brands defining the next era aren't watching it happen from the sidelines.
Meet the Mirakl team at Cannes Lions, June 22–26, for curated conversations, exclusive events, and 1:1 strategy sessions built for where the industry is headed.
Commerce-first media starts in Cannes.
Why nobody funds it
Which brings us to the structural problem, and the reason this is a fitting place to end the series.
The foundational data work doesn't have an obvious owner. In that same Flywheel survey, only 25% of CPGs had a single team leading their agentic commerce strategy at all. So you've got work that's expensive, slow, invisible, and organizationally orphaned.
Anne sees one structural fix starting to emerge: the rise of the chief digital officer. Historically, e-commerce and retail media each sat in their own department and laddered up separately to a CFO or COO. The CDO is the role that finally puts the site experience, e-commerce, and often the media business under one person — which means the infrastructure work that feeds all three finally has somebody accountable for it. (Plenty of these businesses still sit under the CMO, to be clear. The shift is early.)
But here's the part that ties this topic back to retail media. When that foundational work does get funded, it's frequently the retail media network that ends up carrying it — because, as we covered last week, the RMN often has the most modern data stack in the building. So the media team inherits a company-wide infrastructure bill it didn't ask for, to enable an AI future the whole organization will benefit from.
Mirakl's own origin story is a version of this, which is presumably why Anne sees it so clearly. The company spent years building marketplace software, and the recurring problem was gaps in retailers' catalog data — products that couldn't be properly shopped because the attributes weren't there. That was a niche infrastructure headache long before LLMs made it everyone's problem. The boring work was always the work. But AI has raised the stakes.
Now what
I don't have a tidy fix to offer, and Anne didn't pretend to either. The shortcut demon isn't a villain so much as a temptation — the entirely human urge to reach for the fun part and assume the foundation will sort itself out.
It won't. The clone can only find what's been tagged. The AI can only sell what's been described. And the retailers who'll do well in an agentic world are, unglamorously, the ones already in the kitchen.
An RMN's pitch to advertisers is built on its data: who the shopper is, what they bought, what they'll surface for next. Strip out the foundational work and the agentic upside the whole industry is selling has nothing underneath it. The RMNs that do well in an agentic world will be the ones that quietly did the baking while everyone else was reaching for donuts.
The kids reach for the donuts. But somebody still had to bake the cake — and in a lot of these companies, it's the retail media team who's wearing the apron.
That wraps the series. If you want to argue about these demons in person, Anne and I are hosting an IRL Breakfast Club at Cannes — Thursday June 25, from 9am, with RMN leaders from Costco, Albertson's, Tesco, Halfords, CVS, and more. Apply to attend here.
This 3-part series is sponsored by Mirakl Ads.
