The Shift Report · Issue 7 · 28 June 2026
How to choose your first AI project in operations
Most operations don’t lack opportunities to use AI. They lack a clear idea of which jobs AI can actually take on. Start with the recurring irritation, not the transformation on the strategy slide.
It’s 6.15am and it’s getting messy. A driver hasn’t turned up. The transport planner has both monitors full, working out what the impact will be and how to deal with it. He is also trying to get the guy on the phone. Someone else is scrolling back through yesterday’s emails to find the one where the later slot was actually agreed.
Later that morning, in a meeting room, a different group from the same business sits down to talk about its AI strategy.
Most operations do not lack opportunities to use AI. They lack a clear idea of which jobs AI can actually take on. The best starting point is rarely a dramatic transformation. It is usually a recurring irritation: a task that eats time, crosses several systems, and depends too heavily on one person knowing where to look. Your team could name it in a second, even though it never makes the strategy slide.
So don’t begin by discussing your AI strategy. Begin with the work everyone is tired of doing. The recurring irritation is the one task you understand well enough to define, where the cost of getting it wrong is small, and where everyone already agrees it’s broken. That’s the combination an AI project actually needs. The transformation senior management wants to announce usually has none of those three.
Three things worth knowing
1. A robot you can talk to still needs a job worth describing.
At its “Delivering the Future” event in London on 4 June, Amazon unveiled the next version of its Proteus robot, which takes plain-English commands with no programming, inside a plan to put over €10bn into its European operations. Announced alongside it: $1bn to upskill its workforce, fully funded training for roles like robot maintenance and data analysis.
My take: the robot does exactly what you tell it, so what you get back depends entirely on how clearly the task is defined. Plain English makes the machine easy to instruct. It does nothing for a fuzzy process. If you couldn’t describe this job to a new starter on their first morning, a conversational robot won’t rescue it — it’ll just produce the confusion faster. Define the task first.
2. You probably do not need to hire an “AI supply chain expert.”
Gartner research reported on 17 June found demand for people who combine supply-chain experience with AI skills rising far faster than supply. Gartner’s own conclusion is that you cannot close the gap by recruiting. You will have to build the capability inside the building.
My take: that’s good news. The person you need may already know why an apparently simple delivery plan falls apart at 3pm, which maintenance record is usually wrong, which customer instruction needs checking. What they’re short on is the confidence to use AI deliberately and to challenge what it gives back. Most of the time I would choose training my own people who know the work over betting on an AI expert trying to improve work they do not understand.
3. Be careful you don’t automate away how people learn.
The World Economic Forum’s 22 June report on entry-level work looks at what happens when AI absorbs the tasks people used to cut their teeth on. The efficiency is real. So is the risk: you can remove the very work through which your future supervisors build judgement.
My take: some routine jobs are hidden apprenticeships. Preparing a shift report teaches someone what matters. Checking transport paperwork teaches them where errors hide. Hand all of that to a system and you save a few hours while quietly weakening the next set of managers. The fix isn’t to protect pointless admin. Let AI draft the first version, then make the apprentice check the evidence, fix the gaps, and explain the conclusion. Take away the drudgery. Keep the thinking.
One thing to try this fortnight
If you’ve never taken an AI tool past simple search or chat, picture this. You’re back from two weeks off. A delivery failed while you were away, it’s escalated, and there’s a customer meeting in two hours. The people who could brief you are tied up. What you’ve got is a long email chain and a pile of attachments, half relevant and half not.
Open Microsoft Copilot, or whatever AI your company gives you, and type: “I’m getting ready for a customer meeting about a failed delivery. Read the email chain and attachments below and bring me up to speed on everything that matters: what happened, what was promised, what’s still open.” Drop the whole lot in and see what comes back. Then ask follow-up questions the way you’d grill a colleague who’d been covering for you. It will get you meeting-ready in minutes, and the first time it does, the penny drops.
SOPwise turns a Standard Operating Procedure into a training package: task-based steps, critical mistakes, and scenario questions. £35 per SOP, no subscription. Built for UK manufacturing and chemical distribution.