AI Commerce Weekly: Week 28, 2026
Britain is one of the heaviest AI-shopping markets and the most sceptical. The same week, the industry and the FCA agreed the rulebook, not the tech, is the holdup.
TL;DR
W28 built its case in three layers, then went quiet two days early. Attribution came first: M&S put Lily AI behind its product feed, while London's Fleek and Whering and Berlin's reverse.fashion raised money to make used and owned stock machine-readable. Then accountability, named twice in one week. The payments industry at Money20/20 said the agentic-commerce technology is built and the rulebook is not, and the FCA's Mills Review recommended foundations for agentic finance. Under both sat the sharpest UK number of the week: dunnhumby found Britain is one of the heaviest AI-shopping markets and the most sceptical, with 39% not trusting it. Capability was never the constraint. Trust was.
Britain is the most sceptical AI-shopping market
The number to carry out of this week is an awkward one, and it is ours. dunnhumby published European research across eleven markets finding that the UK is simultaneously one of the heaviest users of AI for shopping and the most sceptical market surveyed. Thirty-nine per cent of Britons say they do not trust the technology, the highest scepticism in the study, and yet usage here runs ahead of most of Europe. Germany is the mirror image, pairing high usage with the highest trust, where 33% mostly or completely trust AI. Ireland is approaching twice the European average for AI-assisted grocery activity. So the British position is not low engagement. It is high engagement married to low belief, which is a harder thing to design for than simple reluctance.
The adoption is generational, as you would expect. Twenty-four per cent of 18 to 34-year-olds across Europe have used AI shopping tools, against just 4% of the over-55s. What people actually use it for is narrow and utilitarian, and that is the useful part of the data for anyone planning a roadmap.
Sarah has made the point before that the industry keeps reading adoption off usage when the number that matters is belief, and this is the cleanest illustration of the gap all year. A Briton who has used an AI tool to compare prices has not thereby agreed to let it buy anything. dunnhumby's Chief Data Science Officer Sandra Stanley frames the consequence as a doubling of the job to be done.
That is the sentence that organises the rest of the week. You now have two customers to win. The shopper is won with visible control and a product that does not embarrass itself. The agent is won with clean, connected, well-governed product data, so that when it reasons over your catalogue it can actually find you and describe you correctly. The supply side spent the first half of the week on the second of those two problems.
The supply side makes the garment legible
If an agent is going to do the choosing, the unglamorous precondition is that your product has to be readable by a machine before it can be recommended by one. Four separate moves this week were all, underneath, the same move: making the garment legible.
The clearest, and the most relevant to a UK audience, was Marks & Spencer putting Lily AI's Product Intelligence Platform behind its product feed to generate structured attributes at scale across Google, organic search and emerging AI channels. M&S says the deployment cut the manual effort and the number of external partners it needed to manage product data, while lifting feed quality. Lily AI claims stronger visibility, higher click-through and a meaningful revenue lift, and that is a vendor claim rather than an audited figure, so hold it loosely. The framing from Lily AI's Purva Gupta is the bit worth keeping: in a world where your product feed is your storefront, the quality of your product content is a competitive advantage. That is the attribution thesis in a sentence, and it is why this is a more strategic story than a routine martech deal.
Around M&S, the money moved the same way. London's Fleek raised a 25 million dollar Series B, led by Burda with eBay and others, to scale Fleek Sort, a vision-language model that identifies, grades and prices secondhand garments from images and video so that used stock can be processed digitally rather than by hand. Berlin's reverse.fashion took a seven-figure pre-seed from High-Tech Gründerfonds for AI textile sorting that classifies garments by condition, brand, size and material and routes them to reuse, resale or recycling, with Digital Product Passport integration baked in. And London's Whering raised 7 million dollars, led by eBay Ventures and the Google AI Futures Fund, to expand a wardrobe app that now has more than 10 million users and turns what people already own into structured, styleable data. New stock, used stock, recycled stock, owned stock. Four rounds, one theme. The fashion win this week was upstream in the data layer, out of the shopper's sight, exactly where it was last week when New Look put generative AI in the design studio. If your AI roadmap points only at the shopfront, you are working on the half of the problem the customer can see and ignoring the half the agent reads.
The rulebook, not the technology, is the holdup
Here is what made W28 cohere rather than just accumulate. In the same few days, two very different rooms reached the same conclusion: the hard part of agentic commerce is no longer the engineering.
At Money20/20 Europe, the payments industry said it plainly. Payment Expert's write-up gathered Silverflow, Worldpay and Equals around a single argument: Visa, Mastercard, OpenAI, PayPal and Google have all shipped agentic-payment systems, so the technical pieces are largely in place, and the unresolved work now sits around the transaction rather than inside it. Agent identity, liability in the fault cases, dispute handling and fraud. Silverflow's Robert Kraal put it most usefully.
The specifics are worth knowing because they will land on your roadmap. Equals expects the competing agent standards to shake out to two or three, the usual Betamax versus VHS story, and treats an agent as a party in its own right with its own verified identity. Worldpay argues for mutual recognition between schemes, public agent registries that work the way passports work across borders, rather than two hundred closed registries nobody can reconcile. On liability, the working assumption is that the cost falls on the consumer where the agent acted within its instructions, which is a consumer-protection question waiting to happen. And on fraud, the awkward inversion is that systems built to block every bot now have to let the good bots in.
The same week, the UK regulator arrived at the same door. The FCA published the Mills Review, billed as the first AI-impact review initiated by any financial regulator, and among its seven recommendations is to enable the foundations for agentic finance: a trusted agent protocol, pre-authorised mandates and clear liability rules. It put a demand number under the appetite too, with FCA-commissioned research finding around 11 million UK adults would be likely to use AI acting autonomously within pre-set goals. The Review is about retail financial services, not retail goods, so do not overread it. But when the payments industry and the regulator independently name agent identity and liability as the binding constraint inside one week, that is the signal. The Payments Association's line, that firms should treat agentic AI as an accountability and governance issue now, is the right instruction for a retailer too.
AI adds to search, it does not replace it
A quieter finding this week deserves attention precisely because it is now the second study to say it. dunnhumby framed AI as an addition to existing search rather than an end-to-end shopping solution. Independently, and with a completely different method, a Fevad and University of Toulouse experiment reached the same verdict. Testing 450 people in France and 450 in the US on a washing-machine purchase, the researchers found Google's search engine remained superior to AI systems, both general-purpose and brand-built, throughout the decision. Users of conversational agents asked for more information but were far less likely to reach the point of buying, and confirmed they are not yet ready to pay directly through ChatGPT or Gemini.
Two things in that study should give a technology leader pause. The first is that AI performed better for rational, utility purchases and worse than human advice for experiential ones, which maps neatly onto dunnhumby's price-comparison-first usage pattern and tells you where to point AI first. The second is more pointed: consumers were less inclined to trust brands' and retailers' own proprietary AI. That is an uncomfortable finding at a moment when half the sector is building an in-house assistant, and it argues for earning trust in narrow, verifiable tasks before asking a customer to believe your bespoke bot over the search bar they already know. The agent-only storefront remains a thesis about the future, not a description of the present, and this week produced two independent data points saying so.
Agentic SEO gets productised
If the agent reads your catalogue before it recommends you, someone was always going to sell you a way to check what it sees. This week that someone was Lantern, which launched an "Agentic Commerce Platform" that monitors how products are surfaced and interpreted inside ChatGPT, Gemini and Claude, then applies fixes to product pages and structured data with a human approval step before anything goes live. It is US-led and its market-context figures are its own and unaudited, so file the vendor as a watch-item. But the shape of the thing matters, because it is the same machine-readability thread the feed tracked through geoSurge and Nudge in recent weeks, now packaged with the human-in-the-loop gate that the week's accountability conversation makes look less like a nicety and more like a requirement.
You do not need to buy Lantern to act on the idea, and the cheapest version of the audit came free this week from the payments floor. Equals' advice at Money20/20, echoed in the same Payment Expert feature, was blunt: if you sell white shoes, ask an agent to show you some white shoes and see where you appear. Do that across ChatGPT, Gemini and Claude with your own bestsellers, and treat whatever comes back as your baseline. It costs an afternoon, it needs no procurement, and it tells you whether the attribution work in the section above is a project you have already started or one you have not.
Sources
Britain is the most sceptical AI-shopping market
dunnhumby is a customer-data-science firm; the figures are agency research, attributed to dunnhumby. Base 1,400+ shoppers across 11 markets.
The supply side makes the garment legible
Lily AI's visibility/CTR/revenue-lift claim is vendor-reported, not independently audited. Fleek's platform-scale figures and Whering's and reverse.fashion's uplift figures are company-stated.
- Marks & Spencer enlists AI to sharpen online product discovery (M&S x Lily AI), Retail Gazette, 9 July 2026
- Fleek secures $25m to digitise secondhand fashion (Series B, led by Burda), Tech.eu, 8 July 2026
- Berlin's reverse.fashion bags seven-figure funding to scale textile sorting (HTGF pre-seed), Tech.eu, 8 July 2026
- Whering secures $7m to expand AI-powered wardrobe app (eBay Ventures, Google AI Futures Fund), TheIndustry.fashion, 7 July 2026
The rulebook, not the technology, is the holdup
The Mills Review addresses retail financial services, not retail goods; its agentic-finance recommendation reads across to agentic-commerce governance but is not retail-goods regulation.
- Agentic commerce is built. The rulebook isn't (Money20/20 Europe: Silverflow, Worldpay, Equals), Payment Expert, 10 July 2026
- FCA publishes landmark review into impact of AI on retail financial services (The Mills Review), FCA, 6 July 2026
- AI and the future of retail financial services (The Mills Review, full report), FCA, July 2026
AI adds to search, it does not replace it
The Toulouse study (Trend(s) Chair) is not fully published until September 2026; figures are from the 1 July Fevad Grande Conference unveiling as reported.
Agentic SEO gets productised
Lantern is US-led and its context figures (4,700% YoY AI traffic, 75% of new product searches in LLMs, 5-8x conversion) are vendor-reported and unaudited. Watch-item.
Stay Connected
Follow LLCommerce on LinkedIn
Get the latest AI commerce insights, analysis, and industry news delivered to your feed.
Large Language CommerceAuthors
