Before any AI upskilling programme: are you lifting the bottom or extending the top? Different goals, different programmes, different failure modes. Most organisations never choose.
Every AI capability programme is quietly optimising for one of two things. Raising the floor: bringing the broad middle of the organisation from occasional, hesitant use to confident daily practice. Or raising the ceiling: taking your most advanced people further, into governance, into production, into work that changes what the business can offer.
These sound compatible. In practice they compete for the same budget, the same calendar, and the same executive attention. And they fail differently. A floor programme fails invisibly: everyone attends the training, nothing changes on Tuesday. A ceiling programme fails expensively: impressive pilots, no path to production, a frontier team that becomes an island, and eventually a second company living inside your first one.
Most organisations never make the choice explicitly. So they buy a generic programme that serves neither: too basic for the advanced group, who disengage, and too abstract for the middle, who nod along and revert. The budget gets spent. The board asks what changed. Silence.
Here is the uncomfortable arithmetic. The return usually lives in the floor. Your advanced people will develop with or without a programme. They are already experimenting at home, already pushing tools past their intended use. And there is a quieter truth: some of them have already reached their own ceiling, of capability or of what the organisation allows them to do with it. Every additional training euro spent there buys less than the one before. What this group needs is structure, governance, and a route from experiment to production, which is different work entirely. The middle will not move on its own. It is the largest group, the most underestimated, and the one where a shift in daily behaviour compounds across every process it touches. The ceiling work is real and worth doing, but it is a different scope, with different people, and it deserves its own design.
So before you sign off on any AI upskilling: which one are you funding? What does success observably look like for that group? And who is going to tell your most advanced people that this programme is not for them?
If you can't answer the last question, you're about to buy the generic version.
We design capability programmes around the group with the highest return on behaviour change, which is almost never the group that talks about AI the most.