97% of organisations report no real return on their AI investment that they can actually point to. Kristen Hansen and I agree on the number. We do not fully agree on the cause, and that turned out to be the useful part.
My diagnosis: execution
I have spent two years arguing that AI programmes stall for structural reasons. Nobody owns the outcome. Nothing ships in weeks. Governance gets bolted on at the end rather than designed in. The programmes that reach production share those three things, and the ones that stall are missing at least one.
That remains most of the picture, and it is the half I can fix with a method.
Kristen's diagnosis: the people
Her research says something uncomfortable. 44% of Gen Z employees admit they resist or actively sabotage AI adoption. Not because the tools are bad — because they are afraid of becoming obsolete.
Her explanation is mechanical rather than moral. When a person is in what she calls a threat state, everything narrows: thinking capacity, the ability to learn new skills, the ability to collaborate, the willingness to change. The amygdala has detected a threat and the person becomes, in her words, smaller.
You can build the right agent, into the right workflow, with the right owner — and still watch it fail, because the people using it are in a threat state.
Where the two meet
The implication I had underweighted is that these are not sequential problems. You cannot finish the execution work and then start the human work. A well-built agent arriving in front of a frightened team is already a failed deployment, and no amount of governance fixes it after the fact.
Kristen draws a distinction I have found useful since. Managing AI adoption is process work — rollout plans, training schedules, adoption metrics. Most organisations are reasonably competent at it. Leading AI adoption is something else: motivation, emotion, and the thing that makes someone willing to abandon a way of working they have used for twenty years.
You can mandate the training. You cannot mandate that someone stop being afraid their job disappears in five years.
What actually works
Her prescription is narrower than I expected, and more practical. Create psychological safety around experimentation, with clear guardrails rather than none. Let people share small wins publicly, because that is social proof and it moves a team faster than any mandate. Do something every week that moves forward, instead of running a single campaign of urgency and hoping it carries.
One of our own deployments bears this out in a way I did not plan. Across 500+ people in procurement, category management, sales and revenue growth, adoption reached 95% inside six weeks. A tool people are compelled to use is a compliance problem. A tool that many people choose, that quickly, means it was genuinely faster than the old way — and that is the only thing I have seen reliably beat a threat response. Make the new way obviously better, then let people prove it to each other.
The correction
So the gap is bigger than I have been saying. It is execution and the human capacity to absorb it, which makes it a leadership problem twice over.
That does not change the method. It changes what has to be true before the method works.