For months, our AI conversations at the leadership table went in circles. Anecdotes duelled with anecdotes. One leader had a transformative story, another had a sceptical one, and each meeting relitigated the same question, is this actually working, with fresh energy and no resolution. Everyone was arguing from their corner of the elephant.
What ended it was almost anticlimactic: a single adoption score, reported consistently, month after month.
The score itself was a simple composite, active use, depth of use, spread across teams. I’d be the first to admit it’s imperfect. Any single number describing something as textured as an organisation learning a new way of working is wrong in a dozen ways. I could critique it for an hour.
But here’s what it did that nothing else had managed. It gave everyone the same object to look at. Debate stopped being “my anecdote versus yours” and became “why is that number what it is, and what would move it?” Which teams are ahead, and what are they doing differently? Why did it dip after the restructure? What happened in the area that jumped? The conversations became specific, comparative and forward-looking, which is everything the circular debates were not.
The other thing a consistent number does is accumulate meaning. A single reading tells you little. Eight months of readings tell you whether your program is a story you’re telling yourselves or a change that’s actually happening. Boards, in particular, are starved for this. They hear AI enthusiasm from every management team in every portfolio; a tracked, honest, unflattering-when-it-should-be number is rarer and worth more than any showcase.
One warning from our experience: the moment a number like this becomes a target rather than an instrument, it starts lying to you. We kept it as a thermometer, not a KPI, and I’ve defended that line more than once.
If your AI discussions keep circling, ask what single, consistently reported measure would give the argument somewhere to land. Then have the discipline to report it when it’s unflattering.