Why license utilization tells your board almost nothing
Seat counts and weekly active users are the two most misleading numbers in a bank's AI programme.
6 min read
The number everyone reports
Most AI programme updates open with the same slide: licenses purchased, licenses assigned, weekly active users. It is easy to gather and it moves in the right direction, which is exactly why it survives.
The trouble is that opening a tool once a week is not a workflow change. A banker who pastes a paragraph into an assistant and then rewrites the memo by hand counts as active, and produces no saving at all.
What actually separates the desks
When you look at desks that genuinely got faster, the difference is rarely tooling. It is whether people can judge an output quickly, know the escalation path when something looks wrong, and have seen a colleague do it well.
Those are behaviors, and behaviors can be measured directly rather than inferred from a login record.
A better reporting line
Replace the utilization slide with a readiness distribution: how many people sit in each tier, which dimension is weakest, and which function moved since the last cycle.
It is a harder number to produce and a far easier one to act on, because every tier has a specific intervention attached to it.
Where to start
Pick one function with a clear before and after, baseline it, run a focused intervention on the weakest dimension, and measure again a cycle later.
One credible before and after will do more for your programme's funding than a year of utilization charts.