Keeping honest over what's AI hype and what's a genuine moonshot is an ongoing battle for many at the moment. The time between the release of new models, product, and vendor pitches is getting shorter continually, and this fever pitch is making it hard to act strategically.
We wanted to power a conversation for retail leaders where they could step off the carousel of pitches and think more strategically about what's working. A space to focus on the lived reality and conversation that creates a clearer mental model for what to prioritise, how to make it work, and how to hold a clear strategy that impacts the bottom line.
What follows is the learnings of our recent breakfast roundtable for senior leaders across retail, FMCG, hospitality and consumer goods. A curated group of invited practitioners and leaders talking honestly about where they are.
Current retailer position
The feeling in the room wasn't excitement or fear. People have moved past both. AI capability isn't futuregazing - it's the here and now, and leaders are in the weeds of real organisational and technological change.
That change comes with a common trifecta of pressures:
1. The board above
2. Cross-functional coordination across
3. And POC chaos within.
Each one has its own aspects. Together, they make holding a coherent direction harder than the headlines suggest. There's anxiety about enablement, about people being left behind, about getting caught chasing a mirage of productivity/ROI and ending up worse off than where you started.
Against that backdrop, real wins are emerging. One retailer has replaced a whole legacy BI stack with an LLM running on top of a clean data warehouse: sales questions that used to take a team a weekend now come back in five minutes. Another has cleaned up product data across millions of marketplace lines in ways that would have been impossible eighteen months ago.
But these wins have hard limits. Almost all of them are read-only. A key point made universally was that every "agent" deployed is still a lightweight research assistant, not a co-collaborator or work-to-be-done colleague. The version of AI that writes back into core systems - that can close the loop by acting - sits behind a governance problem no one has solved yet. The block isn't the AI model or its capabilities. It's the lack of monitoring, controls, and nerve to let an agent do something without a human in the way.
Build-versus-buy has shifted in the same direction. The coding agents everyone has access to now, with the right guidance, can build a SaaS replacement with a single prompt and a token spend of pennies (relatively). But possible isn't the same as right. One of our attendees offered his CFO several hundred thousand pounds in savings by cutting a subscription. The offer was declined with this observation: "What happens if you leave the business tomorrow? Who keeps this system you built running?" AI lets one person do the work of a team. But the strategic decisions still need a team to live with them. If you want to go fast, go alone. If you want to go far, go together.
Beneath all of this is an awkward conversation about people. Boards are planning five years out on the assumption that AI cuts headcount. The leaders in the room don't buy it. Their view, near-unanimous, is that you don't cut headcount, you change the mix. Juniors will be different, more seniors may be needed, the costs remain equivalent but the skills mix changes. The headlines about AI writing code, as one person said, are doing real damage to the conversations they now have to have with their executive teams. They don’t understand why one rewrite takes a week instead of months, but another is still a multi-year transformation effort. The core work of transforming and modernising foundational technical and data cores remains.
Where the customer fits
The other thread running through the morning was the customer. Three pictures emerged.
The retailers furthest along - the ones who put the work into platforms and data years ago - are pointing AI at personalisation, supply chain and operations. New ways of working, not new propositions. It's measurable by the business and even where there’s no direct line of sight to customer benefit, it’s easy to prioritise for the efficiency gains.
Some, mostly the digital-native brands, are experimenting aggressively in interesting areas. They're using AI to build customer devotion in new ways. Making individual feel seen by dialing in relevancy. Designing for lifetime value and in-moment value, not next quarter's generic conversion metric.
Then there's agentic shopping. Not a new channel - an entirely new operating model for discovery. A new way customers come to you, if they come to you at all. No one in the room knew what lay ahead, but they worried that discovery, price transparency and opening their data to the machines to consume gives it away with no transaction quid pro quo. It's not hard to see why this emerging threat/opportunity is vexing: it feels noisy, hype-led, and it isn't where the visible conversion gains are coming from. Owned brand agents, like Amazon's Rufus, offer more control over customer experience and paths to transaction. But as we all live with daily, we aren't all Amazon.
Most have landed on the same answer: get your product data and stock clean, make sure an agent can find you, hold the question of transactions for later. But it's a bet, not a plan. Feed your data to the LLMs and you may never see the sale. Don't, and you may not exist in the new channel at all.
What stuck with me most is how much of the brand work retailers have done over the past decade now sits at risk. Unified customer experience. Brand voice as a component of loyalty. The careful work to make you feel different from the shop next door. All of it flattens when an agent picks for the customer without their human attachment.
The leaders who come out of this well will be the ones who work out how to keep their brand alive in a channel they don't own. Distribution is not brand. Those treating it as a metadata problem alone may find themselves in a race to the bottom.
What I'm taking away
The real version of where retail is on AI doesn't sound much like the version peddled by Silicon Valley. The leaders doing this work are running two conversations at once - showing visible progress on the productivity story their boards want, while quietly putting in the foundations, governance and skills that will decide whether any of it changes the business.
The ones I'd back are the ones who can hold both without letting either drown out the other.
We're running another of these in a few months. If any of this lands with you, get in touch.