"We deployed AI, but it hasn't changed much"
I hear it often, and almost always from transformation directors in large groups. The tools are there. The licences are paid. The training has been delivered.
Then I ask the next question: what has actually changed in the teams' work? They search for words. Then silence.
That silence is the subject. This is not a technology problem, it is a practice problem. I saw it very concretely on a project team of about ten people: the tools had been available for months, but were being used alongside the existing work. Nobody had touched the way work flowed.
The pattern that produces this outcome
I have supported dozens of teams on this, and the sequence that fails is always the same:
→ pick the tool
→ deploy it
→ the teams resist
→ blame the teams
The fourth step is the most expensive, because it closes the file with a false explanation. The problem does not sit in the teams' resistance; it sits upstream: a tool was introduced into a workflow nobody had looked at.
I have seen teams adopt AI in 2 weeks and others stall for 6 months. The difference had nothing to do with budget, and nothing to do with the tool.
Start with the frictions, not with the tool
The process I apply systematically starts before any question of tooling, and it comes in three steps.
Map the existing frictions. Before talking about AI, I look at where the work already jams. It is the least spectacular move in the approach and the one that decides everything else, because AI does not repair a broken workflow: it amplifies it. Plugging an assistant into a badly designed step produces the same badly designed step, faster and in greater volume.
Identify the high-return tasks. Not all frictions are equal. The ones that deserve AI are simultaneously repetitive, time-expensive and judgement-poor — precisely the ones nobody claims credit for in an annual review.
Integrate into the working gesture, not next to it. This is where most deployments are lost. An assistant you have to go and fetch in a separate tab will not be used. The one that gets used is already in the tool where the team works, in the ticket, in the channel where they talk.
On the engagements where this sequence was held, I observe an average 40% reduction in cycle time. There is nothing magical about the figure: it comes from the first step, not from the tool selected.
The real problem is not being late
"I'm behind on AI." Just last Thursday, a manager said that to me. He tests ChatGPT, Copilot, Gemini, Claude. So does his team. But nobody is able to say whether any of it actually creates value.
Companies are not short of tools. They are short of results. According to Gartner, only 21% of AI projects generate real value — a figure I meet in the field with depressing regularity.
This is not a maturity problem, it is an approach problem. AI is still used as a productivity tool, to go faster. But going faster in the same place transforms nothing: it produces more of the same thing, including more of what was already useless.
The comparison that distorts everything
The famous lateness also deserves to be put in its place, because it weighs on decisions more than anyone admits.
Out of 8.1 billion humans, the vast majority have never opened an AI tool. A small fraction try a chatbot from time to time, a fraction of that fraction pays for a subscription, and a handful use an assisted-coding tool. If you are reading these lines, you are statistically in the last categories.
In other words: a tiny bubble, permanently comparing itself to early adopters even deeper inside the bubble, and feeling guilty about it. That sense of lateness is an artefact of position, not a diagnosis.
Those who rushed without method are not ahead either. They are managing the debt today — technical, organizational, human. I have described elsewhere the liability generative AI quietly contracts when nobody provisions for it.
The question to ask on Monday
If you keep only one move from all this, it is not to buy, to train or to compare models. It is to take one team and trace the real path of a request, from arrival to delivery, noting every waiting point.
You will then know where AI has a chance of creating value in your organization, and where it will merely accelerate a traffic jam. That is the approach I detail in convincing through usage rather than speeches.
We are not testing AI. We are testing our own capacity to transform the work around it.