Model 10 · AI Measurement Architecture

Every AI programme has a dashboard. Far fewer can answer whether the programme did what it promised.

The problem is not a lack of metrics. It is usually the opposite. Usage, prompts, licences, logins, sentiment, incidents, cost and audit evidence all exist somewhere. They just live in different teams, different tools and different conversations.

That is how a programme can be green everywhere locally and still fail the board-level question: did it work?

Measurement Architecture reframes measurement as infrastructure. It is one KPI tree across four domains: Value, Service, Compliance and Adoption. Each domain is owned by the function or module that already runs it, and every measure rolls up to the North Star success measures set at the beginning.

The discipline is the pairing.

Lag measures tell you whether you arrived. Lead measures tell you whether you still can. A lag without a lead is a verdict you can no longer change. A lead without a lag is activity pretending to be outcome.

The model refuses vanity metrics unless they connect to the promise. It also refuses the four-dashboard sprawl where every team reports its own truth and no one can see the programme.

The central idea is simple: usage is not value, and a dashboard is not proof. Measure the promise, not the platform.