AI programmes are transformation programmes

We tracked 63 AI initiatives to month 18. The ones owned by operating committees delivered a median 68% of their value case; the ones owned by IT delivered 23%. Frequently the technology was identical. What differed was whether anyone who owned an operating decision, such as an order, a price or a service queue, was accountable for the model changing it.

The mechanism is unglamorous. A forecast that no ordering forum consumes is a chart. An answer engine that no service process trusts is a demo. AI creates value at the moment an operating decision changes, and IT does not own operating decisions, through no fault of its own.

The screening test we now apply is one sentence. Which operating decision does this change, and who owns that decision? In a recent engagement it killed 31 of 42 candidate use cases in a week, which was the highest-return week of the programme. The eleven survivors all reached production, because each had an owner whose Monday changed when the model worked. We would not read much into a survival rate that clean. The test is good at removing use cases that were never going to land, and we have no evidence yet that it improves the ones which survive it.

The consequence for structure is equally short. AI belongs on the operating committee agenda, priced per use case and sequenced like any other commitment. That is not a demotion. It is important in exactly the way every other operating change is important, and the machinery that delivers operating change already has a name.