The tool is the last question you ask
Edition 050 · 24 September 2026
A marketing manager walked up to Andrew Afonin and asked for an AI account.
He asked why. She said she had a couple of reports and spreadsheets she wanted to rebuild using AI. He asked why again, and she didn’t have an answer, because nobody had ever asked her to have one. Left alone, she’d have ended up with a second spreadsheet, saying nothing the first one didn’t.
Andrew, Head of Digital Innovation at UK Trade Furnishings, told that story on a panel at Meet Magento UK this year, hosted by JH’s Nick Muldoon alongside Matthew Lawson, Chief Digital Officer at Ribble Cycles. I sat in on it, and the thing that stuck wasn’t really about AI. It was about how many requests for it arrive with nobody having worked out what they’re actually for.
Access to the tool was never the constraint at UK Trade Furnishings. Knowing what to point it at was.
Andrew’s business didn’t plan its AI rollout so much as let it start ad hoc: pockets of the business trying things because someone was curious, months before anyone drew a line connecting them. That’s normal, probably unavoidable in year one. What changed things was the order of operations his team put in place for anything that wanted to graduate from one person’s experiment into something the business actually relied on.
Data first: where does it live, does anyone genuinely know, and how does it move between the systems that don’t talk to each other. Workflow second: pull the process out of the one person’s head currently running it and put it somewhere someone else could follow. Team readiness third, before the tool question even gets asked: will the people who’d use this actually accept a process they didn’t build themselves? .
Skip that order and you get a quieter kind of failure. Andrew’s team built what they call a company gateway: one entry point, department logins, each team with its own slice of the tools. They built it because without one, every department’s enthusiasm for AI was producing its own login, its own tool, its own small island of value that didn’t connect to anything else. Solving the adoption problem created an integration problem, and somebody still had to go and solve that one too.
Matthew’s team at Ribble Cycles started somewhere more predictable: the developers. Within a year, in his telling, nobody on that team had written a line of code by hand. That didn’t shrink the team. It freed up the slack to go looking for the next place AI could actually help.
Customer service came next, and the method mattered more than the tool. Instead of asking the team what to build, Matthew’s developers sat inside the department and worked out what customers were actually contacting them about. Most of the answer lived in one person’s head, and the job was getting it onto paper: a knowledge centre built from what that person already knew, with automated responses layered on top once it existed. Customer service didn’t disappear as a function. Its job changed, from working through the same repeat questions to going and finding the ones nobody had solved yet.
The build that actually shifted adoption across the rest of the business, though, wasn’t the customer service system. It was Claude inside Excel. Ribble’s finance director lives in that spreadsheet and was never going to hand the cash flow over to an agent. But watching Claude work inside the tool she already trusted did something a dev-team success story couldn’t. It made the rest of the business believe AI was theirs to use too, not just the developers’.
Matthew and Andrew landed on the same boundary from two different directions, without comparing notes beforehand. Matthew tells his team that whatever comes out of Claude or ChatGPT is still their work, the same way nobody credits Excel for a spreadsheet it helped build.
Andrew puts it in starker terms: decision-making is the one thing that can never move to the AI, because responsibility can’t move with it. You can’t blame a model for a bad call. Someone still has to own the decision that the data, the workflow and the readiness were supposed to set up.
That part gets skipped in most AI coverage, because it’s less exciting than the tool. The tool gets easier and cheaper every month. The order everything has to happen in before you reach for it doesn’t.
The tool was never the hard part. Knowing what to point it at, and who answers for what it produces, still is.
One more thing. All the talks from Meet Magento UK 2026 are live now on our YouTube channel, so you can watch this one in full along with the rest. We’ve also got an ebook of this year’s sessions on the way, this one included, and you can sign up to get it the moment it lands.
This post was originally written for CEO & Founder, Jamie Huskisson’s LinkedIn newsletter, Breakthrough Commerce.