Docebo surveyed 2,000 employees and learning leaders on AI training. The number everyone quoted is the least informative one in the report.

Written for anyone about to build a training plan around which side of a confidence gap to believe. 77% of managers and 24% of employees are both self-reports. Only one figure in the same report is actually tied to something that happened — and it isn't that one.

Docebo's new survey of 2,000 enterprise employees and learning leaders produced the number every outlet led with: 77% of managers believe their people are set up to use AI well, while only 24% of individual contributors strongly agree. Docebo's own write-up calls this a confidence gap and recommends closing it with training built around real work. That framing assumes one side of the gap is closer to the truth. Neither side's number is evidence of anything close to that.

What the survey actually asked

Docebo's "AI Readiness Gap" report, released April 7, 2026 and fielded by the research firm Centiment, surveyed 2,000 people across the US, UK, Canada, France, Germany and Italy — 1,000 individual contributors in non-HR/L&D roles and 1,000 learning leaders (CPOs, CLOs, HR and L&D managers and above). The headline split comes from the same underlying question put to two different audiences: learning leaders were asked whether their organization has set employees up to use AI well, and individual contributors were asked whether they agree. 77% of the leaders said yes. 24% of the employees strongly agreed. A third figure in the same report is narrower and more useful: 85% of employees say the training they actually received hasn't helped them use AI in their actual role, and 56% say they're too buried in manual work to find time to learn the tools meant to lighten it.

Two self-reports, zero evidence

Measuring what changed matters more than measuring how anyone feels about it, and that applies before you even get to which side of this survey is "right." A manager's belief that their team is trained is a self-report of confidence, not a measurement of anyone's ability to do anything with AI. An employee's disagreement is also a self-report — of a feeling, not a skills test. Nobody in either group was asked to attempt a task and have the result checked. A 53-point spread between two unverified guesses isn't proof that managers are complacent or that employees are struggling; it's proof that both groups were guessing, in different directions, with nothing behind either guess. That's a reason to stop trusting both numbers, not a reason to average them or side with whichever one sounds more cautious.

The one number in the report worth keeping

The 85% figure is more useful, and not because it's any less self-reported — it is — but because it's a narrower claim. "I think my team is ready" and "I don't feel ready" are judgments about an entire capability. "The training I received didn't help me use AI in my actual role" is a report about a specific, remembered attempt: a session happened, someone tried to apply it afterward, and it didn't transfer. That still isn't proof the training failed on its own terms — a bad memory, a missing tool, or an unrelated blocker could explain it too — but it's tied to something that either did or didn't happen in someone's real work, which puts it well ahead of a confidence score about the organization as a whole.

The diagnosis the report never ran

Both headline numbers describe disagreement. Disagreement isn't a diagnosis. A program built to close a 53-point confidence gap is solving the wrong problem if the actual constraint on a given team is something the survey never asked about directly — and the report's own 56%-overwhelmed figure points at exactly that kind of constraint: people too buried in manual work to go practice the tool that's supposed to replace some of it. Training aimed at making managers and employees agree with each other closes the gap between two guesses. Training aimed at finding out what's actually stopping a specific team — time, access, trust, or a tool nobody's shown them how to use on their own work — starts by setting both guesses aside and asking what the team genuinely can't yet do.

A test neither number survives

The fix isn't a better survey question that gets closer to the truth on the next attempt; every version of "do you feel ready" still measures a feeling. The fix is a workflow that actually gets built and that someone checks on weeks later: a named task, a named owner, a point in the process where someone reviews what came out, and a date to check whether the thing is still running under a real deadline. That's falsifiable in a way neither of Docebo's numbers are. The workflow either exists and is still in use in six weeks, or it isn't — and finding out doesn't require taking anyone's confidence, or anyone's doubt, at their word. The before/after measurement that proves it is the same one that should have run before either survey question got asked.

None of this is a defense of the status quo Docebo's report is pointing at — 85% of people saying their training didn't transfer is a real problem, and 56% saying they have no time to learn the tool is a related, concrete one. The argument is narrower: the number that led the coverage, the 77-versus-24 split, is the least informative figure in the report, because it's the only one that measures how two different groups feel about each other's competence instead of what either group actually did.

This is a position, not a critique of Docebo's survey design, which is a standard methodology credible for what it measures. The 85% and 56% figures are the report's own, more specific findings; this piece argues the 77/24 confidence split that led most coverage of the report is the one number in it worth setting aside.