Wouter Neef
5 min read
June 5, 2025
Most leadership teams we meet have already accepted the argument for data and AI skills. The budget conversation is not the hard one anymore. The hard one is why a program everybody agreed to still has a completion rate in the teens six weeks in.
The honest answer is that engagement, not content, is where these programs are won or lost. A well-built course with no momentum behind it performs worse than an average course with a community, a leaderboard and someone noticing when you fall behind. And the drop-off is rarely at the end — it is between module one and module two, which means the fix is almost always in the first fortnight.
What moves it: weekly challenges with something at stake, a visible leaderboard, drop-in sessions where people bring their own problem, a channel where questions get answered by peers rather than a helpdesk, and a nudge to the individuals who stall rather than a broadcast to everyone. None of that is glamorous. All of it is the difference between a program and a library.
Cohort-based learning works, and we build most programs that way, because a shared start date and a shared group create the social pressure that self-paced learning lacks. But it is a tool with conditions attached, and it is worth being clear about them.
Cohorts earn their keep when the subject is new, when there is genuine resistance to overcome, or when the goal is behaviour change rather than tool proficiency. They struggle when the population is very large, spread across many time zones, or when people simply need one specific skill at one specific moment. We have seen a client move from fixed cohorts to a self-paced model, replacing kickoffs and graduations with booster sessions and office hours — and for their situation it was the right call.
The mistake is not choosing one or the other. It is choosing without knowing which of those conditions you are actually in.
Skill assessments are the most common way leaders try to prove value, and the most common place the proof quietly falls apart.
When participants score highly on the baseline assessment before the program has even started, the instinct is to celebrate. It is usually a sign the assessment is too easy. A baseline with no headroom cannot demonstrate growth, and a small rise from an already-high starting point is indistinguishable from noise. We have had to revisit our own assessment difficulty for exactly this reason.
Build the baseline to discriminate. Ask the same questions every cohort so the numbers stay comparable. And put self-reported confidence next to something observable — what managers see differently in the work — because those two do not always agree, and the gap between them is the interesting part.
The last one surprises people. These programs succeed or fail on ownership, and the right owner is often not the data organization.
One large client put it plainly: the majority of their transformation cases turn out to be HR cases. Who gets trained, how it fits a career path, what happens to roles as the work changes, how you get people to show up — those are people questions, and a data team is not equipped to answer them. The programs that stall tend to be the ones owned by a function with all of the subject-matter expertise and none of the levers.
There is a second ownership question underneath it: data ownership. If nobody is accountable for the quality and the definitions of the data people are being trained to use, the training teaches them to distrust it faster. Governance and literacy are not sequential projects competing for the same budget. Run them together, or run the second one twice.
If you want to talk through which of these applies to your organization, get in touch.
On 24 September we are hosting the 10th Data & AI Literacy Round Table in Utrecht: HR and data leaders in one room, on who owns AI literacy and what that asks of leadership. Invite-only, 30 to 40 seats.
Member of
Nominated
