Why AI Training Isn't Changing Behaviour

We ran the training and by every visible measure it went fine. Everyone showed up, many participated, a few people even seemed excited, asking deeper questions. A month later the majority had returned to doing everything the way they always had, with an expensive AI subscription sitting unused like a gym membership in February.

If you have rolled out an AI tool, some version of this is probably familiar. It is tempting to read it as a training problem, to assume people just need another session or a better prompt guide. That instinct is the trap. More training rarely fixes it, because training was never the thing standing in the way.

Training was never the bottleneck

There is a figure that gets quoted constantly, that only about ten percent of what people learn in training ever shows up as changed behaviour on the job. I would not lean on the exact number, because its origins are shaky and researchers have challenged it for years. What does hold up is less catchy and more useful: whether training changes behaviour depends far more on the environment people return to than on the training itself.

Knowing how to do something and actually doing it under pressure are two different problems. A workshop can solve the first. It does almost nothing for the second, because the second is governed by everything waiting for people back at their desks: the deadlines that never moved, and the quiet sense that the old way of operating is still faster right now. AI is the most vivid example of this gap I have seen in years, precisely because the tools are so capable and the behaviour change is so hard.

The numbers behind the stall

The scale of it is striking once you look. McKinsey’s 2025 research found that while 88% of organizations report using AI in at least one function, only around 6% report significant enterprise-wide impact from it. Usage is nearly universal. Value is rare. The same research found that only about a fifth of companies using generative AI had actually redesigned any of their workflows around it, which turns out to be the factor most associated with seeing real returns.

A separate study from MIT, looking at the state of AI in business in 2025, put it even more bluntly: the large majority of organizations were seeing no measurable return on their generative-AI spending, while a small minority captured almost all of the value. I want to be careful with that finding, because it often gets repeated as “95% of AI projects fail,” which is not what it says. It says most organizations are not yet getting measurable financial impact, which is a statement about value capture, not about the technology breaking.

Adoption is a behaviour, not a briefing

Put those together and a pattern emerges that has very little to do with the tools. Organizations are buying access and calling it adoption. They are training people on features and expecting changed behaviour. The gap between those two things is where most of the disappointment lives.

This is, I would argue, the same problem behind almost every stalled change effort, just wearing a newer and shinier outfit. Adoption is not a knowledge state you reach by being shown the buttons. It is a set of behaviours people actually repeat, under real pressure, until the new way becomes the path of least resistance. A briefing cannot install that, only a changed environment can.

Where it actually turns

So what moves the needle is rarely more training. It is redesigning the actual work so the AI sits in the path of how things get done rather than off to the side in another tab. It helps when leaders model the behaviour themselves instead of delegating it downward. And it depends on giving people enough ownership of the change to push through the awkward early phase where the old way still feels faster.

If that sounds familiar, it should. It is the same gap I wrote about in why strategic plans fail, and it turns on the same thing as every other change: whether people own the new behaviour or are merely complying with it, which I get into in how to get real buy-in for change.

Training has its place. It just cannot carry adoption, engagement and impact on its own, and treating it as if it can is why so many AI rollouts quietly stall. Closing that gap, the distance between a tool people have and a behaviour people use, is the work my services are built around.

So if your AI rollout has gone quiet, the question is probably not who needs more training. It is what would have to change about the work itself for the new behaviour to become the easy one.