We asked eight retailers to be honest about AI

Earlier this month, we hosted our first retail customer advisory board in London. Merchandising, trading, and technology leaders from eight retailers spent a day with us, comparing notes on AI, data and how decisions really get made. The day ran under the Chatham House Rule, so nothing here is attributed to anyone in the room. But the themes were consistent enough that they're worth sharing.

I went in expecting a conversation about capability: models, agents, what the technology can and can't do. But what we got was a conversation about people. Across every single session, adoption and change management came up far more often than any product, platform, or feature.

Adoption beats capability

Every retailer in the room named adoption as one of their biggest challenges. Some are working with processes that are 20 or 30 years old, shaped by years of teams working in isolation. One now has a dedicated change management team just to keep things moving.

Others described a quieter barrier. People tend to focus on what a new tool can't do, rather than what it can. AI still feels relatively foreign to a lot of teams, which makes education every bit as important as rollout. And rolling out a system isn't the same as using it: more than one business has modern planning tools in place while much of the real work still happens in spreadsheets.

The fix the room kept coming back to was simple: start small. Get one thing working, make the win visible, and let it spread. One retailer described starting with an internal “Ask HR” agent, learning from it, then scaling department by department. The ambition is a single pane of glass, where people use AI without having to think about it.

Until you get some success doing something a new way, you carry on the other way.

That line, from one of our attendees, sums up the day better than anything I could write. Capability gets you in the room, but adoption is what keeps you there.

Connected isn't the same as in control

We opened the afternoon with findings from The Reaction Gap, the UiPath survey of 500 UK retail leaders conducted earlier this year. Ninety-eight percent describe their systems and teams as connected during peak trading. Only a quarter have the real-time visibility to act on issues as they happen. And more than two thirds only respond once a problem has already hit commercial performance.

The room recognized that gap immediately. The hardest problem isn't a lack of data, but getting the numbers to match across back-end systems, when every department manipulates the same data in its own way and the same figure comes out differently depending on who you ask. Several retailers run a war room during peak season, with various dashboards lighting up every screen. The challenge is the “so what”: hundreds of data streams, and very little clarity on which three issues matter right now.

That's why data foundations came up in almost every session, from merchandising to the shop front. Poor or siloed data blocks agentic use cases before they start. One retailer is working to make every piece of product data intent-based, so customers reach what they need as fast as possible. Speed to the customer starts with data you can trust.

Keep people in the decision

We asked one table a simple question: if agents did the legwork on every trading decision, how would your merchandising team change over the next three years? The answers were less about replacing people and more about reshaping their day.

The model several retailers described was clear. Reduce big decisions to a yes or no. Let agents prepare the call and clear the routine ones, perhaps 90% of them, and let people approve the 10% that need judgment. But that only works if the system explains itself; every recommendation needs to show its reasoning and the risk of acting on it, and teams need the data literacy to read the dashboards in front of them.

Then came the tension nobody had fully solved. Today, merchandisers learn to make good calls by doing the manual pre-work: building the Monday trade pack, pulling the numbers, spotting what looks wrong. If agents take that away, how does the next generation build the same judgement? Would a graduate go straight to making yes or no decisions without ever learning why?

There's no easy answer, but it's the right question. Building AI into a team means designing how people keep learning, not just deciding what they stop doing.

"I need this to work"

Our Reaction Gap research found that 97% of UK retailers have implemented AI in some form, yet 47% are still waiting to see any commercial impact from it. Nobody in the room was surprised. Part of the gap is simply lead time. A returns-prediction model, for example, only proves itself once a range has been sold and sent back.

But the bigger issue is that ROI is hard to prove and expected anyway. Retailers want to know whether the commercial value of a new technology is worth what it costs, and which metrics will show it. Often, you have to invest first just to build the case.

One attendee offered a measure I keep thinking about. Forget the business case for a moment; the clearest sign that something is working is when the people using it say, “I need this to work.” Demand from the team is the proof point that matters most, and it's the one a small, visible win creates.

Where to start: the work nobody is prioritizing

If the advice is to start small, the obvious question is where. The room had a clear answer: the middle band of work that sits below the headline transformation programs. It rarely gets a budget line of its own, but it's where a huge amount of time disappears.

Non-stock procure-to-pay came up as a prime example. Supplier onboarding can take up to eight weeks, and orders raised before it's complete create exceptions further down the line. Teams spend their days chasing sign-off, managing amendments after delivery and working across information silos. A meaningful share of non-stock orders may not be needed at all, which means AI could help reduce spend, not just optimize it. What retailers want is a single source of truth, a view of what's stuck at each stage, and no more manual purchase order creation.

That's exactly the kind of win the room described: contained, measurable and visible to the people doing the work. The technology is ready, and the retailers who win won't be the ones with the most AI. They'll be the ones who make it stick.

We'll be bringing the advisory board back together in April. In the meantime, my thanks to everyone who joined us in London and gave their time so openly. The best ideas we heard came from you.