RTS 2025: Where UK Retail AI Came to Work

The move from Olympia to ExCeL was the clearest possible statement about where this show is going. Olympia is a beautiful Victorian space — built in 1886, and carrying every operational implication that implies. ExCeL is a purpose-built exhibition centre with a logistics ethos. It scales. With 400+ technology suppliers and upwards of 15,000 senior retail professionals over two days on 2 and 3 April, the move felt overdue rather than ambitious.

What made RTS 2025 different was not the new venue. It was the quality of the AI conversation.

The Deployment Shift

A year ago at Olympia, AI was everywhere and shallow. Every second exhibitor had something with "AI-powered" in the strapline. The main stage sessions were heavy on potential, light on evidence. This year at ExCeL, something had shifted.

The conversations I kept having, both in sessions and on the floor, were about deployment: what got built, what was tested, what the numbers looked like after six months. Demand forecasting tools that had been put through the stress of the supply-chain disruption coming out of US trade policy changes, and which had given meaningful lead time over manual scenario planning. AI-assisted returns triage that was shortening processing time and beginning to accumulate the predictive data that makes returns prevention possible. Workforce scheduling optimisation that had been running long enough to have a full seasonal cycle behind it.

None of these are the applications that generate keynote headlines. They are the ones that show up in the P&L. That shift, from potential to evidence, is what a maturing technology conversation looks like. British retail does not rush at new things, which means when it does commit, the case has usually been made.

The Ecommerce Programme

The ecommerce content at RTS 2025, curated with Commerce Futures, was the part of the programme I spent most of my time in. The topics ranged from AI-powered product search and customer data infrastructure to headless commerce architecture and the emerging agentic commerce question. The coherence was better than previous years, and the sessions attracted the kind of audience that actually works on these problems, not just thinks about them.

The Google AI Mode implications for product discovery were getting serious attention. Practitioners were starting to see the traffic impact in their own analytics. Not catastrophic yet, but visible enough to be a planning concern. The conversation about optimising for AI discovery systems rather than pure SEO was being had by people who were actively implementing changes, not just conceptualising them.

One session that generated substantial follow-up conversation in the networking spaces was a panel on AI-generated product content at scale. The quality variance problem was being confronted directly: the gap between "AI can generate product descriptions quickly" and "AI generates accurate product descriptions reliably" is a real operational gap. Even a small error rate across tens of thousands of SKUs adds up fast. Retailers who had deployed at scale were learning this the hard way.

The Honest Mid-Market Picture

The session I found most valuable was the one framed explicitly around the mid-market challenge. What do you actually do when you are a £20-200M UK retailer, your AI options are real, your resources are limited, and every case study you hear is from a company ten times your size?

The panel's answer was honest rather than diplomatically hedged. The highest-ROI AI applications for mid-market UK retailers right now are not the frontier ones. They are email personalisation (platforms like Klaviyo and Ometria are producing measurable uplift at this scale), returns management automation, and customer service triage. The foundation-building work — getting product data in order, unifying customer data, making site content machine-readable — is what makes the more sophisticated applications possible later. That infrastructure work is unglamorous. It is also the most important investment a mid-market retailer can make in 2025.

The message was not what anyone comes to an exhibition to hear. It does not look impressive in a vendor keynote. But it is the message that practitioners with limited capital and limited technical resource can actually act on.

The Mood

RTS 2025 felt like an industry that had done some of the hard early experimentation and was beginning to separate the applications that work from the ones that overpromised. The British retail sector does not rush, and it does not forget previous hype cycles. The caution is structural. But it was paired at this show with something more concrete than I have seen in previous years: a clearer shared sense of where the value is and, as importantly, where it is not.

A year ago I wrote that the gap between enterprise AI capability and mid-market access was the defining tension at this event. It still is. But it is beginning to close. More slowly than the vendor pitches suggest, and faster than the most sceptical voices expected.

That is roughly where you want to be.