Most organisations approach AI upskilling as a catalogue problem. They identify the skills the organisation needs, find programmes that cover those skills, and deploy them. When this does not produce the expected results, they add more programmes.
The reason it does not work is not the quality of the programmes. It is the diagnosis. The real problem is distribution — AI readiness is not uniform across the organisation, and uniform programmes applied to non-uniform populations produce systematically uneven results.
In a typical enterprise of 5,000 people, AI readiness is distributed roughly as follows: a small proportion — often 5 to 10 percent — are already active AI users who are self-educating and experimenting. A larger group — perhaps 30 to 40 percent — have awareness and curiosity but no structured exposure. The majority have some awareness but significant anxiety, scepticism, or both.
Within each of these groups, the distribution varies further by function, level, and geography. The data science team and the finance team have almost nothing in common in terms of AI readiness, even if they are at the same level of the organisation.
When you run a single AI programme across the organisation, it will be calibrated to some target learner profile. It will be too advanced for the bottom quartile of readiness, appropriate for the middle, and too elementary for the top.
The top quartile disengages. They learn nothing and leave the programme more sceptical of L&D-driven AI initiatives than before. The bottom quartile is overwhelmed. They leave with fragments of knowledge they cannot apply and often a reinforced sense that AI is too complex for them.
The middle quartile — the largest group and the most important to move — learns something, but without the context to know how to apply it, and without the support structure to practice it.
A programme calibrated to the average of a wide distribution helps no one well and some people actively badly.
The first step is measurement, not programme design. Before deciding what to teach, organisations need to know where their people actually are. This is not a mood survey — it is a structured skills audit that maps actual AI interaction patterns and capability levels across functions.
The second step is segmentation. Different populations need different programmes — not just different content, but different intensity, pacing, format, and expected outcomes. The data science team needs a fundamentally different AI programme than the sales team. Trying to serve both with the same curriculum is a false economy.
The third step — and the one most organisations skip — is supporting the high-readiness population differently. This group does not need programmes. They need communities of practice, access to advanced resources, and the organisational permission to experiment and share. Putting them through another introductory programme is worse than doing nothing — it signals that the organisation does not understand what they already know.
Distribution-aware AI upskilling requires a platform that can serve different populations at different levels simultaneously — tracking individual progress, surfacing the right content at the right time, and giving managers and L&D teams visibility into where capability gaps actually are rather than where they are assumed to be.
Without that infrastructure, even the best-designed multi-track curriculum becomes difficult to manage and measure. The infrastructure question is not a technology decision — it is a programme design decision that needs to happen before the content decisions, not after them.
Browse our upcoming batches — live, instructor-led, delivered on Orbit.