Banks Love a Pilot. Then Nothing.

Banks love a pilot. Six weeks, one department, a vendor in the room, a demo at the end that everybody claps for. Then nothing. The pilot was never the hard part. Any competent vendor can make a model look good on a clean slice of historical data. The hard part is the thing nobody scheduled: getting that model into a process somebody owns, with a number attached to it, and a decision about what happens when it is wrong. That last one is where most of them die. Ask who is accountable when the model declines a good customer, and the room goes quiet. Not because nobody cares. Because it was never assigned, and assigning it means somebody has to accept risk they did not have last month. So the pilot gets extended. Then it gets a phase two. Then it quietly becomes a slide in an annual review about innovation. The institutions getting real value are doing something duller. They pick one process with an owner who already carries a target. They put the model inside that process. They measure the same number they were measuring before and see whether it moved. If you are about to approve another pilot, ask a different question first. Which process will this live in, who owns it, and what number does it have to move by when? If nobody in the room can answer that, you are funding a demo. *Stanislaus Martins trains enterprise teams across Sub Saharan Africa on practical AI adoption. [Speaking and training](/speaking).*