The work, not the tool: why AI transformation fails

TL;DR: AI transformations fail the way all transformations fail, in the operating model rather than the strategy deck. With AI the specific failure is handing the whole thing to the people who understand the tool and calling it adoption; Klarna did that, IKEA didn't, and the difference was the operating model, not the technology.

The people who understand the work are the first experts in any company, and tool expertise is most useful when it's added to theirs. So the one move that's right whatever your AI strategy turns out to be is to get the people who know the work using the tool on the tasks they already own, and to check whose job has actually changed.

A couple of weeks ago I gave a talk on why AI transformations fail. It’s a live topic and the answer is still emerging, so I focused on anecdotes that bring the main failure points to life. The talk led to some good conversations, and those have helped me sharpen the thinking, though it’s certainly still evolving, as the whole space is. The short version: the companies getting AI right start from the work, not the tool. The rest of this is why that’s harder than it sounds, and what to do about it.

The claim itself is old. Transformation doesn't fail in the strategy deck, it fails in the operating model, and by operating model I don't mean the boxes and lines. I mean the one that actually runs: who decides, what the company can actually deliver, how people work day to day. I was making the same argument about digital in 2019, when the research we used said about one in eight transformations delivered what the deck promised, and the decks were good.

What's specific to AI is the shape of the failure. The sales pitch is tool-led: buy the tool and you'll need fewer people. So the business, reasonably enough, hands AI to the people who understand the tool and waits to see what comes back. There's a vendor, a budget and a pilot, so it looks like adoption, but it's not really.

Two stories make the point better than I can. Klarna, February 2024: their AI assistant handled 2.3 million customer conversations in its first month, the work of about 700 people, roughly 40 million dollars a year. It was the most quoted AI slide of that year and it was true. Fifteen months later the CEO said publicly that they had focused too much on cost, quality had suffered, and they were hiring people back. Credit to him for saying so. The technology worked; the operating model was a cost line.

IKEA put a chatbot on customer service in 2021, same technology, same job. It took the routine calls, more each year. The interesting part is what happened to the people on the phones: around 8,500 of them were retrained over a couple of years as remote interior designers, and that channel now does over a billion euros a year. They started from the work, a customer who wants help with a room, and let the bot take the part of it that was routine.

Same tool, opposite answers. That's the main point: the work, not the tool.

James Allen runs a great series in which he interviews CEOs about how they've handled AI, and he went in expecting to hear a familiar story: I knew I was out of my depth, so I hired the right people and got out of the way, and heard none of it. What he heard was closer to what Michelle Zatlyn of Cloudflare told him: even when we're not the experts, we show up, we ask the questions and we learn, so that we have better instincts. His conclusion was that deferring to the experts is a form of abdication.

I'd add that it isn't just CEOs. The same thing happens to leaders deeper in the organisation, the heads of marketing, operations or sales who are closest to the work. I'm starting to see it in my conversations through my work at IEDC – Bled School of Management, which is itself rethinking what executive education should look like when AI sits inside every job, and which puts me in front of leaders working through exactly this. They hand AI to the AI people because that's who seems qualified, and they step back.

I think part of the problem is how we use the word expert. We've started using it for the people who know the tool. But the first experts in any company are the people who know the work: how a customer is won, where the time goes, what a good output looks like and what a bad one costs. Tool expertise is valuable when it's added to that. On its own it knows what the tool can do and has no idea what the company is for. It's also easy to see why the tool people end up leading. They're confident, the work people are busy, and AI doesn't arrive as a project with an owner and a budget line; it arrives inside existing jobs, one task at a time, so nobody owns it.

The obvious objection is that the work people don't know what's possible, and that's true, which is why the tool people belong in the room. But knowing what's possible is the easy half. Knowing what's worth doing is the half the business can't outsource. And someone at the top still has to decide, and keep deciding, what gets protected while the work changes underneath it.

My working view, and I'd be glad to be argued out of it, is that this is easier to get right with AI than it was with digital. You can teach a good work person enough of the tool in a few weeks. I've never seen anyone teach a tool person a business, and I've watched it attempted. The thing that compounds is a team that knows both, and only one of the two is hard to come by.

So if I had to pick one thing, one move that's right whatever your AI strategy turns out to be: get the people who know the work, and know what the company is there for, learning the tool and experimenting with it on the tasks they already own. They're the only ones who can tell a good output from a bad one, and the only ones who can put it into the work. The platform, the vendor and the roadmap will all change; that team stays.

The question for your own company is simple. Who is leading the work on AI: the people who understand the tool, or the people who understand the work, with the tool people alongside them? The easiest check is to ask whose job has actually changed in the last six months, not disappeared, changed. If the answer is nobody's, the tool people are leading. This post grew out of my talk at Game Changer Zagreb 5.0 in October. Thanks to the organisers and to everyone who engaged with me afterwards.