Does a course listing prove anyone can use AI at work?
Your dashboard shows 2,400 employees enrolled through the company's new AI skills marketplace, and the CFO wants to know what that number actually proves. It looks like momentum. It might just be a long list of saved courses nobody finished.
Common practice treats marketplace enrollment as an adoption metric: the more courses listed and seats filled, the more progress you can report upward. That's fair as far as it goes - a marketplace does widen access to AI courses across roles, teams and locations. We think a growing catalog is too often mistaken for a growing skill base, when it only measures what's on offer, not what anyone absorbed.
Worth noting up front: none of the research below studies AI skills marketplaces or skills fests as a format directly. It's evidence about general training access and completion, applied here by analogy to that narrower question.
- Enrollment shows who had access to a course, not whether they opened it
- Completion shows someone reached the end, not whether it changed their work
- A full catalog shows breadth of offerings, not depth of skill across the workforce
But access isn't skill. Docebo's 2026 AI Readiness Gap Report, surveying 1,000 learners at enterprises with 1,000-plus employees across six countries, found that 1 in 5 had received no AI training at all, marketplace or otherwise. A catalog sitting unused proves nothing about whether people can apply AI to their actual work.
JFF's survey of 2,754 US workers, fielded in November 2024, found only 31% reported receiving employer-provided AI training in the first place.
The same JFF survey found most people learning AI on their own rather than through any employer program, with 60% citing self-directed learning as how they actually picked it up. People are building skill outside your AI skills marketplace whether you track it or not.
Sources: AI Readiness Gap Report 2026 | Docebo, https://info.jff.org/hubfs/JFF%20AI%20for%20Workers%20&%20Learners%20Survey.pdf
Does attending an AI skills fest change what someone can do?
Picture a two-day AI skills fest in 2026 where 300 employees rotate through vendor booths, lunchtime demos and a closing keynote on "the future of work." Attendance badges get scanned, a wrap-up email goes out, and L&D logs the headcount as a win.
That's not nothing. A fest puts AI in front of people who'd otherwise never open a tool, and a good one turns an abstract mandate into something hands-on. We're skeptical, though, that a single event moves anyone past curiosity into real capability.
PwC's 2026 Global Workforce Hopes and Fears Survey, covering nearly 50,000 workers across 48 countries, found only 51% said they had access to the learning resources they need, down from 59% the year before. Exposure at an event doesn't fix an access problem that's getting worse, not better.
The same PwC survey found that most of the 'engine room' - the 56% of the workforce doing the core operational work - said they had never used AI at all, even as adoption rose elsewhere in the business. A packed skills fest and a mostly untouched core workforce can coexist.
Sources: Companies risk losing their most AI-savvy employees while leaving majority behind | PwC, Global Workforce Hopes and Fears Survey 2026 | PwC
Why does finishing training not predict real AI proficiency?
Study.com's 2026 survey of 1,000 US employees, fielded March 16, 2026, found only 18% self-reported consistently producing high-quality AI output with little or no editing.
That's a self-reported outcome, not an independently verified one, so treat it as what employees believe about their own output rather than proof of what a reviewer would actually find.
Study.com's own analysis of that data found a 16-percentage-point gap between how confident people felt about using AI and how often they said their output was genuinely high quality.
We'd treat a completion certificate as an attendance record, not a skill certificate - it tells you someone sat through the material, not that the material changed how they work.
Sources: State of AI Jobs and Skills Report 2026: The Training Gap Slowing Down the AI Revolution - Study.com
Can you trust an employee's own confidence as a skill signal?
That gap between feeling confident and actually producing strong AI output is the real problem with using self-rated confidence as your main skill signal. Confidence tells you how someone feels walking into a task, not what they'll hand you at the end of it. The same pattern shows up above the individual level too: plenty of experimentation, little change in how the organization actually operates.
“Most organizations have high levels of individual experimentation (people using AI tools to complete isolated tasks)”
— Markus Bernhardt, Principal, Endeavor Intelligence · Source: AI Readiness Gap Report 2026 | Docebo
Our contrarian take: asking "how confident are you with AI?" is often the wrong first question. We'd ask what someone actually did with it last week - which tasks, which tool, what the output looked like - before asking how they felt about it.
| Signal | What it shows | What it misses |
|---|---|---|
| Enrollment | Someone had access to a course | Whether they finished it or applied anything from it |
| Attendance | Someone showed up to training or a fest | Whether the session changed how they work |
| Self-rated confidence | How sure someone feels using AI | How good the output actually is, per Study.com's 2026 survey |
None of these signals is useless. Each just answers a narrower question than L&D teams tend to assume it does.
Sources: State of AI Jobs and Skills Report 2026: The Training Gap Slowing Down the AI Revolution - Study.com
What should a head of L&D track instead of enrollment numbers?
Stop grading the catalog. Start grading the gap between what people say and what they can show. Treat AI skills marketplace listings and skills-fest attendance as proof of access and exposure, not proof of capability.
AI Adoption Score takes this approach directly: instead of one enrollment count, it runs nine predefined survey types covering skills, usage, confidence, barriers and more, so you can see where the real gaps sit.
One of those, the AI Individual Skills Distribution survey, maps each person's expertise so training budget gets allocated by the actual gap, not by who signed up first.
A marketplace tells you what's on the shelf. A skills survey tells you who actually knows how to use what's on it - and that's the number worth taking to the board.
Frequently asked questions
Who should own tracking these AI skill signals, L&D or managers?
Either can own it, but L&D is usually the practical starting point since it already runs the marketplace and training calendar. Managers should still be asked for input on role relevance, since they see day-to-day AI use that a central team won't.
How often should we re-survey employees on AI confidence and skill?
A reasonable cadence is every six to twelve months, enough time for training or tooling changes to show up without survey fatigue setting in. PwC's 2026 survey found access perceptions moved by eight points in a single year, so change can happen faster than annual reviews assume.
Is an AI skills marketplace still worth running if it doesn't prove proficiency?
Yes, as long as you stop treating it as proof. A marketplace is still useful for giving employees access and choice; just pair it with a direct skill and confidence survey if you need to report on actual proficiency.
Should we check skill at the individual level or just the team level?
Start at the individual level if you want to allocate training by actual gap rather than guesswork, then roll results up to team and department views for reporting. A team-only average can hide wide swings between people doing the same job.
Sources: Companies risk losing their most AI-savvy employees while leaving majority behind | PwC
