How to Train Your Team on AI: Why the Course Isn't Working
Here’s a number that should bother anyone who has just signed off on an AI training budget. 85% of employees say the training they receive doesn’t help them understand or use AI in their role.
The obvious reading is that companies aren’t training people. That reading is wrong, and it’s worth being precise about why — because the fix you reach for depends entirely on which problem you think you have.
Companies are training. It isn’t landing.
SHRM surveyed 5,875 US workers this spring and found that nearly 40% said their workplace had run workshops on practical, day-to-day AI skills. Another 37% got an introduction to the tools. About a third had access to coaching.
That is not a company that’s ignoring the problem. That’s a company that booked the session, filled the room, and got a The percentage of enrolled people who finished a course. It’s the easiest training number to report upward and the least connected to whether anyone can now do anything they couldn’t before. to show for it.
Gallup’s number runs the other way — 47% of employees who use AI say their organization has offered them no training at all — but the two aren’t in conflict. Roughly half of workplaces are doing something. The half that are doing something are mostly not getting their money back.
So this isn’t a volume problem. Running the course twice won’t fix a course that doesn’t transfer.
The thing that actually moves usage isn’t a course
This is the finding I’d put on a slide if I only got one.
Gallup found that employees whose manager actively supports their AI use are using it frequently at a rate of 79%. Where the manager doesn’t, that figure is 46%.
Sit with the size of that gap. It is larger than the gap between having training and not having training. Whatever your Learning and Development — the team that buys and runs corporate training. In a smaller company this might be one person in HR, or nobody. team buys, the variable that predicts whether your team actually uses AI next month is you.
That’s uncomfortable, because a course is something you can delegate and a manager’s ongoing attention isn’t. But it’s also good news: it’s the one input you fully control, and it costs nothing but calendar.
Why generic training doesn’t transfer
The Docebo research that produced the 85% figure also found that 57% of learners say training isn’t very relevant to their specific role. That’s the mechanism.
Generic AI training teaches the tool. It shows you the chat box, the file upload, maybe prompting tips. Then it ends, and you go back to a job made of specific recurring things — the Thursday reconcile, the intake summary, the reply queue — and nothing in the session told you which of those to point this at.
The gap between “I understand what this does” and “I have changed how I do my Thursday” is the whole problem. Training people on the tool and hoping they’ll close that gap themselves is asking them to do the hard half unassisted, on top of their actual workload.
Notice the shape of what does work in SHRM’s data. The most impactful adoption tactics workers reported weren’t one-off events — multiple training sessions ranked among the top three, at 63%. Repetition over time beat the single well-produced session.
What to do instead
The reframe is small and it changes everything: stop training people on AI and start training them on their own recurring task.
Concretely, and in this order:
- Find the task before you book the room. Ask everyone what part of their week they dread that’s basically the same every time. You are looking for repetitive, low-stakes, and checkable.
- Get two or three people through it first, not the whole team. When you eventually run the session, you want to open with before-and-after examples from colleagues the room knows, not a vendor’s demo.
- Make the session hands-on with their own work. Everyone builds one real thing from their own job and shows it. If nobody leaves with something they’ll use on Monday, it was a talk, not training.
- Put something after it. Thirty minutes of office hours a week for a month. Attendance dropping to zero is the correct outcome — it means it stuck.
- Be visibly a user yourself. Given that 79%-versus-46% gap, this is not a soft factor. It’s the intervention.
If you want the full sequence written out — the ground rules to settle first, the pilot, the workshop agenda, and what to do after — that’s what the manager’s AI toolkit is. The order matters more than any individual piece of it.
One caveat worth stating plainly
The 85% figure comes from Docebo, a company that sells corporate learning software. A vendor publishing research that concludes corporate learning is broken has an obvious interest in what you conclude next.
I’m citing it because the finding is consistent with SHRM’s and Gallup’s independent numbers, and because the direction it points — toward relevance and repetition — is the opposite of “buy a bigger content library.” But you should know whose survey it is. Their sample was 2,000 people at enterprises over 1,000 employees across six countries, split evenly between learners and learning leaders, and they don’t publish field dates.
The takeaway
Your team’s AI training probably didn’t fail because it was bad training. It failed because it taught a tool to people who needed help with a task, and then ended.
You don’t need a better course. You need to pick one recurring job your team actually does, get a couple of people through it, show the rest what happened, and then keep showing up for a month. The evidence says the manager is the variable — which means the thing most likely to make this work is the cheapest thing on the list.
Sources: SHRM, “Navigating AI in the Workplace: 2026” — 5,875 US-based workers surveyed March–April 2026; some findings drawn from a randomly assigned subsample of 2,030. Gallup, “AI and Workplace Productivity: What Leaders Need to Know in 2026” — for the 79%/46% manager-support gap and the 47% no-training figure; Gallup doesn’t publish sample size on that page. Docebo, “The AI Readiness Gap” — for the 85% and 57% figures, with the vendor caveat above; the 85% appears in Docebo’s own write-up rather than the ungated report page. The five-step recommendation is mine, not any of theirs.
Related: How to Learn AI at Work When Nobody’s Training You is the individual’s side of this — what to do when you’re the one nobody trained. AI Workshop in Three Hours: You Won’t Get a Day and a Half is the session itself, compressed to what a real calendar allows.