10 February 2026 · 6 min read
The pilot that cannot fail cannot succeed
A pilot without a defined success measure has no way to end. It cannot be declared a success, so it cannot be scaled. It cannot be declared a failure, so it cannot be stopped. It simply continues, absorbing attention and budget.
Before any experiment starts, write down the business measure it is intended to move, the current value of that measure, and the threshold at which you would fund the next stage.
Technical success is not business success
A model can perform well on its own terms while changing nothing about how work gets done. The gap is usually the workflow around it: who acts on the output, what they do differently, and whether that action is actually faster or better than before.
Design the workflow change at the same time as the capability. If you cannot describe what someone will do differently on Monday morning, the pilot is not ready.
Sequence by leverage, not by enthusiasm
The most enthusiastic team is not always sitting on the highest-leverage opportunity. Score candidates on business impact, effort, risk and time to value, and be willing to say no to interesting work that does not move the business.