AI adoption owned by the operating committee and priced per use case. In our data the reporting line predicts AI value better than the technology does, so the practice starts with ownership rather than with models, platforms or another proof of concept. Prediction is not proof of cause, and we say as much on the page where we report it.
A year of pilots has produced demonstrations, a platform bill, and no verified operating result.
AI sits with IT while the decisions it should be changing, in ordering, pricing and service, sit somewhere else entirely.
The board wants an AI position; what exists is a tool list and a risk register.
One or two use cases genuinely work, and nobody can say why those, or how to repeat it.
Every candidate priced against the operating decision it changes, each with a value case, a baseline and a kill test. A use case that cannot name the decision does not get funded.
AI moved to the operating committee, with decision rights and adoption accountability written down. The committee owns adoption, not IT.
Use cases drawn in phases like any other commitment. Three or fewer at a time, verified before the next draw.
Per-use-case delivery measured against month-0 baselines, under the same standard as everything else we ship.
Candidate use cases traced to the operating decision each would change. No named decision, no funding.
Surviving use cases get value cases and baselines. Platform and data work is costed inside them, never as a separate programme with a business case of its own, because a data platform holding its own case will always find a reason to grow.
Ownership is installed at the operating committee. Adoption obligations are named function by function.
The first use cases are released into production under oversight and verified at a fixed cadence. Then the next draw.