Most AI training builds literacy, not adoption. What we see is a consistent gap between a well-run training day and any real change in how people work.
Picture the session. A vendor or consultant runs an engaging afternoon. People try the tool on a sample task, ask good questions, leave with a login and a one-page cheat sheet. Everyone agrees it went well. Three weeks later, almost nobody is using it, and nobody quite knows why, because on paper the training worked.
It worked as training. It didn't work as change. Those are different exercises, and most organisations only budget for the first one.
McKinsey's own internal research on this is blunt about the gap. Across organisations, AI literacy, teaching people what the tools are and how they work, gets most of the investment because it's visible and simple to measure: attendance, completion, a quiz score. AI adoption, actually embedding the tool into how a team's work gets done, is messier, harder to measure, and requires changing incentives and processes rather than running a session. Most companies invest heavily in the first and barely touch the second, which is exactly backwards if the goal is behaviour change rather than a completed training log.
The scale of the problem is well documented outside AI too. Harvard Business Review has called it "the great training robbery": American companies alone spent an estimated $160 billion a year on training, with most of it failing to translate into better organisational performance, because people revert to their old ways of working once the session ends. A separate HBR piece on the topic makes the same point more bluntly: training is the default response to almost any organisational problem, delivered and then left unmeasured for whether it actually changed anything.
Here's the pattern worth naming plainly. A training session changes what someone knows how to do. It rarely changes what they actually do, because knowing how to use a tool and having a reason to change a habit are different things. People are busy, existing workflows are familiar, and a new tool competes with the comfort of the old way, especially when nobody has removed the old way as an option or rewarded the new one.
We'd put it like this: if you handed a team a working AI tool tomorrow and did nothing else, most of them would use it for a week out of curiosity and then quietly drift back to their old process, because their manager never asked about it again, their performance review still measures the old output, and the old spreadsheet still works fine for getting through the day. That's not resistance. It's a completely rational response to an environment where nothing else changed.
The organisations that get a different outcome tend to share three habits, and none of them is "better training content."
The distinction that matters is between a tool people have to remember to open and a tool that's already sitting inside the task they're doing anyway. One consulting firm's own account of this approach describes pairing a coach with a live delivery team for a sprint or two, working through a real task such as claims intake, rather than running a generic session disconnected from the work. Confidence and use rose together, because the learning happened inside the task, not before it.
That's a more expensive way to run training, and it's also the version that changes behaviour, because it removes the extra step of remembering to apply what you learned somewhere else later.
If a manager's team is still evaluated purely on the old output metric, a new tool is optional effort with no visible payoff. Organisations that see real adoption tend to change what they measure alongside the training: rewarding experimentation, tracking use case development, adjusting efficiency targets to reflect the tool being used properly. When the incentive changes, the behaviour usually follows it far more reliably than any training slide does.
Employees watch what their manager actually does, not what the company announces. If a manager tells a team to use a new AI tool and then keeps doing the old process themselves in every meeting, the message that lands is that the new tool is optional. Where we've seen adoption actually take hold, managers were using the tool visibly and talking about it in normal work conversations, not just in the training kickoff.
McKinsey's research on this gives it a number worth remembering: successful AI transformations tend to follow a roughly 1:3:5 pattern, for every dollar spent on the technology itself, three go to redesigning processes around it and five go to capability building and adoption. Most companies invert that entirely, pouring the bulk of attention and budget into the technology and treating behaviour change as an afterthought. The same research found that high-performing organisations were three times more likely to say their senior leaders visibly demonstrated ownership of AI initiatives in their own day-to-day work, not just in a kickoff message.
When a leadership team says "we ran the training and nothing changed," the honest diagnosis is rarely that the training was poorly delivered. It's usually that training was asked to do a job that belongs to workflow redesign and incentive design instead. A training session can teach someone to use a tool in twenty minutes. It cannot, by itself, make using that tool the path of least resistance in their actual day, and that's the thing that determines whether it sticks.
This is why we treat training as one part of an adoption plan rather than the plan itself. The plan needs an answer to what changes about how the team is measured, what the manager will visibly do differently, and where in the existing workflow the tool sits, before the training day is worth scheduling at all.
This lines up with what the World Economic Forum has flagged about skills development more broadly: immersive, real-world experiences close the gap between theory and application in a way a scenario-based workshop alone doesn't, because the practice happens on work that actually matters rather than a hypothetical case.
If your last training didn't stick, the fix usually isn't a better training. Start by asking what specifically will look different in the workflow after the tool is introduced, not just what people will know. Identify who the manager is that needs to model the new behaviour visibly, and make sure that person is genuinely using the tool themselves before anyone else is asked to. Change one measurable thing about how the team's output is evaluated so the new way of working has a visible payoff, even a small one. And put the learning inside a real task the person already has to do that week, rather than a hypothetical example designed for a training room.
None of this requires more content or a longer session. It's close to how we run our AI workshops: working through a team's real tasks rather than a tour of features, so it requires treating the training as the smallest part of a larger change, not the change itself.
Why doesn't AI training change how people actually work?
Because training changes what someone knows how to do, not what they're incentivised or reminded to do differently day to day. Without a change to workflows, measurement, or manager behaviour, people tend to drift back to familiar habits within weeks.
Is the problem the quality of the training content?
Rarely. Most training sessions cover the tool adequately. The gap is almost always what happens after the session, not what happened during it.
How long does it typically take for AI training to fail to stick?
In our experience, the drop-off is visible within about three to four weeks, once the novelty of a new tool wears off and the old workflow remains the path of least resistance.
Should training happen before or during the actual work?
Embedding coaching or support directly inside real tasks tends to produce stronger and longer-lasting adoption than classroom-style sessions held separately from the work, because it removes the extra step of applying a lesson later.
What's the single biggest lever for making AI training stick?
Changing what gets measured or rewarded alongside the training. Without an incentive tied to the new way of working, a training session competes with a comfortable old habit and usually loses.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic +