Wouter Neef
7 min read
June 5, 2025
Europe's competitiveness debate has a recurring conclusion: the bottleneck is not access to technology, it is the rate at which working adults pick up new skills. Participation in adult learning across the EU sits well below the ambition set for it, and it varies widely between member states. The gap is not evenly spread either: it is widest exactly where data and AI are changing the work fastest.
The word usually missing from that debate is speed. A data and AI leader in global pharma put it to us more sharply than any policy paper has: "I don't think it's about adoption, I think it's about the speed of adoption. And humans are at the moment not fast enough to adopt it." That is a different problem from a skills shortage. Left alone, most organizations would get there eventually. The question is whether they get there before the decisions that depended on it have already been made.
For an individual organization, the EU-level number matters less than your own. The practical recommendation is the useful part: become more responsive to changing skill needs, and collect reliable, granular, comparable information about where the gaps actually sit. Without that, upskilling budget goes to whoever asks loudest rather than to where it changes the work.
"Granular" is doing real work in that sentence. An organization-wide average tells you nothing you can act on. The number that changes a decision is the one per team and per role: which group reads a dashboard confidently but cannot question what went into it, which group builds reports nobody opens, which group signs off on AI initiatives without being able to judge them. Those are four different problems with four different answers, which is why we sort people into four data & AI personas before designing anything.
"Comparable" matters just as much. A measurement that only compares you to yourself tells you whether you moved, not whether you moved enough. We assess skills on a 0 to 300 scale before, during and after a program, per persona and per team, and benchmark that against other organizations, so the answer to "are we doing well?" stops being a matter of opinion.
A common response is to dust off the e-learning built a few years ago and bolt on a GenAI module. It rarely survives contact with reality, for a simple reason: the content ages faster than the platform it sits on, and a course nobody finishes teaches nobody anything.
The size of that second problem is easy to underestimate. Completion on standard corporate e-learning catalogues sits around 12%. Put differently, roughly seven of every eight licences you pay for produce nothing at all. Any budget conversation that starts at cost per seat rather than cost per finished learner is measuring the wrong thing.
There is a positioning trap in it too. In nearly every large organization we speak to, the entry level is already occupied: a mandatory, group-wide, largely text-based AI course that everyone has clicked through once. Buying more of that changes nothing, and the people who most need to move have already ticked the box. The tier that is genuinely missing is the one above it, where people apply the material in the tools they use, on the problems they own.
Format is not a detail here either. Carnegie Mellon research puts interactive practice at roughly six times more effective than video-only learning. That ratio is the difference between a curriculum that produces awareness and one that produces work people can do the next morning.
So curricula need to be evaluated on effectiveness and revised, not extended. That means knowing your completion rates rather than your enrolment numbers, knowing which modules people abandon and where, and being willing to cut material that no longer earns its place.
In our own programs the difference between a course people finish and one they quit is almost never the subject matter. It is whether the exercises use their data, their tools and their cases, and whether anyone notices when someone stalls halfway through. Both are deliberate design choices: courses built on the client's own use cases, and someone actively watching progress and nudging instead of a dashboard nobody opens. At Jaarbeurs that combination produced 96% completion across 350 learners; at Skyscanner, 77% across more than 1,400.
Europe has long trailed the US and parts of Asia on productivity, and the gap is clearest in how quickly new technology actually gets used rather than merely bought. The return is there for organizations that cross that line: Morgan Stanley put the net productivity gain at 11.5% for companies with a year or more of real AI use, measured across the five sectors most exposed to AI. Note the condition in that sentence. It is a figure for companies that actually use the technology, not a promise that buying it pays off.
The same asymmetry shows up in what the skills are worth. PwC's Global AI Jobs Barometer put the wage premium for AI skills at 56%, while Lightcast, whose job-posting data PwC draws on, puts it at 28% on a larger sample. Somewhere between a quarter and a half, then, depending on whose sample you take, and rising on both. The labour market is already pricing the capability your program is meant to build.
What those averages hide is how far apart comparable companies now are. In our own conversations with Dutch enterprises of similar size, within a few weeks we heard "a couple of thousand AI agents in production", "fifty thousand", and "zero, because we do not dare yet". That distance is not explained by ambition, and it is not explained by budget. It is explained by whether enough people in the organization understand the technology well enough for anyone to be given permission to act.
There is a human obstacle worth naming. A large share of European workers are anxious about AI, and a notable proportion would prefer government restrictions to protect jobs. You do not resolve that with a compliance module. You resolve it by giving people enough hands-on skill that AI stops feeling like something happening to them and starts feeling like something they can direct.
That anxiety is also why a purely voluntary program tends to reach the people who need it least. The colleagues who sign up first are usually the ones already comfortable. The ones the business case depends on are the ones quietly hoping it blows over.
And the gap is not only at the front line. A data consultancy describing its own client base put it at the top of the house: leaders and board members who are not literate enough to see the consequences of the decisions they take, in organizations that assume the problem is handled because there is a data team somewhere in IT. A program that skips the leadership layer tends to stall one level below it.
Three things follow for anyone running a program. Do not limit it to the data organization. Do not limit it to the people who are already comfortable. And accept that some groups, people who are not at a desk all day, or whose first language is not the language of your training, need a different format rather than a louder invitation.
None of this needs a three-year plan. It needs one cohort and an honest measurement around it.
It helps to agree what "able" means before you measure it. Ours, arrived at while building the assessment: understanding what data is, feeling confident enough to work with it, and applying it in your day to day. Only when all three are true does anything change. Two out of three is a course completion.
Then measure first, so you have a baseline to argue with later. Pick the part of the business where data or AI is visibly changing the work, rather than the part that is most enthusiastic about it. Run one cohort end to end on real cases from that part of the business, then measure the same skills again. After that, train a handful of champions inside the group, so the second cohort does not depend on you.
That sequence is what turns a training budget into a capability the organization owns. The Jaarbeurs program is a worked example: a company-wide ambition, translated into cohorts on their own material, with skill growth measured before and after rather than assumed.
Europe's skill gap will not be closed by any single employer. But the part of it sitting inside your own organization is measurable, and it is the only part you can do something about this year.
Member of
Nominated
