In 2013, John Lewis took a Keane song, handed it to Lily Allen, slowed it right down and sold Christmas with it. A bear who’d never seen Christmas, a hare with an alarm clock, a nation in bits. It worked because it was the familiar thing, done gently, with nobody asked to change anything. Are you humming this song now?
That is also the most accurate description I can offer of UK ecommerce’s growth plan for 2026.
Last Wednesday at ExCeL, Andy Mulcahy Strategy and Insights Director, IMRG opened IMRG’s Growth Summit under the banner Turning AI Into Measurable Growth. His deck had three kinds of data in it: IMRG’s benchmarking of what retailers are actually seeing, live polls of the retailers in the room, and what attendees said they wanted to learn when they registered. Andy shares this type of insight weekly on LinkedIn, I suggest you listen and read. He is also, a healthy skeptic of shiny new things. Given his world of data, this is a useful barometer. The set up for the summit was intriguing too, an old style set dance where everyone got to dance with everyone. Ably chaperoned by Andy himself. Conversation was deep and dark at times, and people in the room had the confidence to speak their mind when asked. Andy, knowing them, held them to task.
The day was dotted with workable, easy to digest insights and he shared that deck with on last week. I have some of my own thoughts.
THE BIGGEST CHANGES ARE THE ONES NOBODY WILL SEE
Asked which part of the online journey AI will change most, 76% of the room said product discovery. On-site search got 9%. Product pages got 4%.
Discovery is the visible bit, so fair enough. But every agent built so far has been designed around a human experience: a human browsing, a human comparing, a human hovering over the blue one. Agents don’t need any of that. They need structured data, clean attributes and something they can parse at speed. Semantic search already rewards pages that describe what a product is rather than how it feels, and metaobjects threaded through the product page are quietly turning the PDP from a shop window into a spec sheet. Four per cent think the product page is on the menu. I’d take the other side of that bet all day.
Andy’s own site search numbers hint at something shifting underneath. Search results bounce rate has climbed from under 10% at the end of 2024 to over 13% in Q2 2026. One reading, and it’s only one, is that shoppers are arriving with their minds already made up somewhere else, and your on-site search is the second opinion nobody asked for.
Ben Thompson would put it bluntly. In an agent-mediated world, your product page stops being a page for a person and becomes an API for someone else’s aggregator. The retailers polishing PDPs for human eyes are building beautifully for the last aggregator. This is a key distinction on why, this will likely be wrong in time. Our frame of thinking is based on designed systems that repeat the stages that we think through. In a similar way it is why ads dont always appear in the way we anticipate. Google has 100’s of signals it looks at for PMAX ads to determine the next best placement, we are asked for our budgets. Even before this evolution we were providing a narrow set of inputs to connect with google signals to try get us the output we wanted - ad performance.
The changes that matter most won’t show up in a screenshot. That’s precisely why they’re dangerous.
THE CHECKOUT ISN’T DYING. BUT IT’S BAGS ARE PACKED.
A real surprise If I am honest. But I did ask myself is the reason , it is dependable - if it ain’t broke then don’t fix it. Checkout also scored 4%. Which is odd, because the most radical version of agentic commerce doesn’t have a checkout at all. There’s a shared token, an acceptance, and that’s it. No basket page, no form, no ‘are you sure?’ No friction. Friction is needed for humans IMO. We beed a healthy level of it. There was use cases here agentic works, and plenty where it does not.
Jen Wright of Rascal Ventures posed the question, later the same day, on a slide of its own: who owns the customer when the purchase experience is no longer yours to manage? Nobody in the checkout poll seemed to have heard it yet. But then, who owns it now. Shopify claim the best checkout out there and unless you are on Plus, there is nothing you can do only implement it.
In fairness to the 4%, the evidence of 2026 is on their side, for now. As we know back in May, OpenAI quietly shelved Instant Checkout in March, and Walmart found conversion inside ChatGPT roughly three times lower than sending people to its own site. The moment money moves is exactly where shoppers want a bit of friction. Some friction is load-bearing. Not all friction is created equally.
The cross pond POV. Strong Customer Authentication under PSD2, still sitting in UK law, means delegated agent payments carry more authentication friction here than in the States. The ‘token and a nod’ flow is a US-first reality. But if we are to grow, my contention is cross border, warts and all, is needed.
So the 4% aren’t wrong that their checkout page won’t change much. They’re wrong about whose it’ll be. Between Google and Shopify’s UCP, OpenAI and Stripe’s ACP, and the card networks’ agent programmes, checkout isn’t disappearing. It’s being relocated to infrastructure the retailer doesn’t own.
ZERO PER CENT. NOT A TYPO.
Loyalty and retention: 0%. Post-purchase support: also 0%. Not one retailer in the room thinks AI changes what happens after you’ve got the customer.
At this point, I put my pint down. On the registration form, Retention & Loyalty was the fifth most requested topic of the day, ahead of checkout, social commerce and creative AI. The same people desperate to learn about loyalty are certain AI won’t touch it. And asked which metric AI will improve most, 44% said conversion rate. Repeat purchase rate got 4%. Average order value got nothing at all. (editing now =: A thought occurred to me that many agencies are now selling loyalty and this may be in part why it is not spoken about. It is talked to, not about).
I mentioned in May in a post, Points Make Permission, Google’s Universal Cart was built on Wallet precisely because Wallet already holds your loyalty cards. UCP added identity linking so shoppers keep their benefits when buying through Google’s surfaces. Sephora launched in ChatGPT with its loyalty programme attached. The platforms have clocked something the room hasn’t: loyalty is the carrot that gets a wary shopper to say yes to the agent the first time. Found money. Member price. No card number to remember.Cheapest token they will ever get.
Meanwhile the SaaS loyalty platforms have spent years turning points schemes into first-party data engines, and first-party data is the only currency that survives an agent standing between you and your customer. Zero per cent says we’re dramatically underrating what those platforms already do, and dramatically underrating how hard Big Tech is working to sit in the middle of them.
If Google thinks your loyalty programme is worth intermediating and you think AI won’t touch it, one of you is wrong. It isn’t Google.
THE WEATHER BEAT THE ROBOTS
Andy’s own summary slide didn’t mince words: in 2026, growth was driven by heatwaves, not AI. His revenue chart shows it plainly. Online sales fell about 10% year on year in April, then spiked into positive territory through May, June and July as the sun came out and Britain bought paddling pools. The slide also carried a line every AI keynote should have to display by law: don’t get too carried away with it.
There’s something gloriously Bill Bryson about it. A nation that has spent a century talking about the weather finally found a use for it: it’s the most reliable growth lever in UK retail.
The AI numbers themselves are sobering. AI assistants account for 0.39% of revenue on average, and first-visit traffic from AI barely registers against search engines at 43%. AI assistant traffic converts at 2.3%, respectable but below email, paid search and affiliates, and only just ahead of organic search. Meanwhile 44% of the room bet on conversion rate as AI’s big win, when IMRG’s conversion chart has sat stubbornly around 3% every year since 2022. I cannot stress this enough. This was my Eureka moment. Doing the same thing repeatedly and expecting different results, is well, mad. Maybe this is where the agentic opportunity lies, maybe it is where all opportunity lies.
Then the channel data, where paid is very much doing the heavy lifting. Organic search’s share of traffic has fallen from around 21% to 18% in a year. Paid search now delivers 14.7% of revenue from 9.3% of traffic, and paid search, shopping and social together bring in roughly 23% of revenue, all creeping up. Organic Growth & SEO was the third most requested topic on the registration form. Retailers can feel the free traffic draining away, and they’re paying to top it back up.
Matt Stoller would point out that the CMA designated Google with strategic market status in general search last October, under the new DMCC regime. The regulator has formally decided Google is a gatekeeper. The industry has formally decided it’s the growth plan.
THE PRIORITIES LIST THAT ARGUES WITH ITSELF
The registration form tells you what’s on retailers’ minds. Top of the pile:
Personalisation & Recommendations (174),
Analytics & Forecasting (153),
then Organic Growth & SEO and Acquisition Channels tied (150),
with Retention & Loyalty (143)
and Performance Marketing (125) not far behind.
Sensible. Familiar. Measurable.
Now look at where AI has actually delivered. Asked where they’d seen the best ROI so far, 54% said written content generation, 41% chatbots, 39% visual content. One-to-one personalisation, the single most popular topic of the day, came in at 6%. We want AI to know our customers. So far it’s mostly writing our product descriptions.
Down the bottom of the interest list sat Cross-Border (70), Retail Media (70) and Customer Support & Experience (71), with Marketplaces (79) barely clear of them. These aren’t nice-to-haves. They are the net-new-money playgrounds. Retail media is the highest-margin line on the P&L of any retailer who’s bothered to build it. But it is not without its own set of challenges. Marketplaces are demand you don’t have to acquire one click at a time. You know my position on marketplaces. Yada, tada yada. Hurrah for ebay. Move. on. Cross-border is, by definition, customers you don’t currently have.
The strongest argument against me, and it’s a good one: ignoring cross-border might be rational. Brexit made EU trade painful. The US ended de minimis for everyone last August. The EU is moving to a flat duty on low-value parcels. Another one, yep. Cross-border got materially harder in eighteen months, and plenty of UK operators have the scar tissue to prove it.
But that’s not strategy. And many of the scars, were of their time. There is gold in tham there hills - there is money to be made if you want to look for it. . Merchant-of-record partners and marketplaces exist precisely to absorb that friction. And from where I’m sitting in Kerry, cross-border is just a Tuesday. Irish brands have always treated the UK as an export market; there’s no domestic 67 million to fall back on. The muscle UK retail let waste after 2016 is the one Irish retail has been using the whole time.Necessity means we have to. In todays PEST climate, we need necessity - mother of invention, right?
We are defaulting to the things we know. Somewhere only we know, if you like. It’s a lovely song. It’s a terrible growth strategy.
THE PEOPLE WHO ARE SURE THEY’RE SAFE
And then the final poll. Respondents picked one role most impacted by AI and one least. 42% picked the Ecommerce Director as least impacted. Not a single person picked it as most impacted. Customer Service topped the most impacted list at 26%, with content managers and merchandisers next at 17% each.
Consider who fills in these polls. Asking a room of senior ecommerce people whether the ecommerce director is safe is asking turkeys to vote on Christmas, and, keeping with the John Lewis theme, the turkeys have voted for a lovely Christmas. (and let’s hope it is, we need little wins).
Here’s the kicker. One slide earlier, 42% of the same room said the retailers who win with AI will be those that use it to support people rather than replace them. Then they pointed at customer service, the one function customers actually touch, and said that’s where the axe lands. We are planning to automate the relationship and protect the org chart. In an agentic world where the customer increasingly never visits you, the few moments of genuine human contact will be worth more, not less.
The most exposed role in UK ecommerce may be the one that feels the most comfortable. Squeaky bum time as Alex Feruson might say.
SO WHAT DOES THIS SAY ABOUT OUR MINDSET?
Read the deck as one document and a pattern emerges. UK ecommerce expects AI to change the parts it can see, and it’s defending the parts it controls. It credits growth to the weather, pays the platforms for the rest, wants to learn about loyalty but doesn’t think AI will touch it, ranks its net-new-money channels near the bottom, and the people answering believe their own jobs are the safest in the building.
That’s not a technology problem. It’s a culture problem, the comfortable kind that feels like prudence right up until it doesn’t. If you run a UK brand, four things worth doing before peak:
Audit your product data for machines, not people. If an agent can’t read your attributes, it can’t recommend you.
Map your loyalty programme to the agent protocols. Decide now whether you want Google holding your members’ benefits, because the default is yes.
Price one cross-border lane properly. One market, one partner, real numbers. Not a strategy deck.
Ask your leadership team who is actually safe. Then ask someone who didn’t write the answer.
WHERE THIS GOES NEXT
Watch three things into 2027: how UK regulators treat delegated agent payments under SCA; whether UCP-style loyalty linking lands this side of the Atlantic, and on whose terms; and whether peak 2026 is once again won by the weather. If it is, we’ll know the robots still haven’t turned up for work.
Jen Wright left the room with four questions for an honest conversation. The fourth was the best: what am I not seeing? On the evidence of this deck, quite a lot. Somewhere only we know is a lovely place to visit. It’s a terrible place to grow.
Thanks to Andy and the IMRG team for the data, and to the retailers in the room for being honest enough to give it. That’s the hard bit. The rest is on us.