AI transformations rarely fail on the technology

In this line of work you hear more failure stories than success stories. That's no surprise — successes each find their own path, while failures look strikingly alike. Put the stories side by side and a pattern emerges: of the most common ways an AI transformation dies, almost none has anything to do with technology. The models get stronger every year. What jams companies is something else.

Death one: hand it to the IT department

It sounds logical: AI is technology, technology belongs to IT. But an AI transformation changes the business — who does what, what counts as good work, how the process runs. IT can build the stage; the business has to perform on it. Transformations led by IT tend to end the same way: a system that is technically impeccable, commercially unused, and quietly deleted at next year's budget meeting.

Death two: the eternal pilot

A three-month pilot goes well. And then? Another three-month pilot.

Pilots are safe: no appraisal has to change, no process has to move, nobody gets upset. But the whole point of a transformation is to make the new way the default way — and defaults require someone to decide, and to own the consequences. Without that person, piloting becomes a substitute for transforming: it looks continuously busy, and permanently changes nothing.

Death three: the incentives fight the transformation

On stage, leadership encourages everyone to use AI. On the appraisal form, performance is still counted in orders processed and emails answered. Employees aren't fools: handing repetitive work to AI means cutting their own numbers with their own hands. When slogans fight incentives, the incentives win every time. Wherever the transformation goes, the appraisal system has to follow — otherwise you are asking staff to choose between obedience and their bonus.

Death four: the boss never touches it

Demanding that the whole company embrace AI while reading one's own reports only on paper. Change draws its credibility not from town halls but from demonstration. When the boss starts using AI — even just to prepare for meetings — the entire company's seriousness about the subject changes overnight. It cuts the other way too: one distracted, going-through-the-motions endorsement from the top cancels ten training sessions.

Death five: the saved time evaporates

The stealthiest one. AI genuinely saves time — reports draft faster, emails clear faster — but nobody decides where the freed hours should go, so they are quietly absorbed by longer meetings, more chatter and "let me think about it". Three months later the review says efficiency is up and the results are exactly where they were. Saving time is only the first half of the game. Where the saved time is reinvested decides whether there is a second half at all.

One root under five deaths

Five ways to die, one root: treating AI as a thing.

Things can be bought, signed off, written into the annual report. But what an AI transformation actually replaces is a way of working — who does what, what counts as good, who answers when it goes wrong. A way of working cannot be purchased. It can only grow, and what it grows on is management's sustained attention, not the sum on the procurement contract.

Technology is never the bottleneck. The bottleneck is whether anyone is willing to own the change.

The good news: all five deaths are preventable, and prevention costs far less than failing once. The hard part isn't knowing — it's answering three questions on the day the project starts. Who owns this? How will appraisals change? What will we do with the time we save? Answer those and the transformation is half done. The other half is where the technology finally comes in.