Pilot purgatory
The failure mode in enterprise AI isn’t the pilot that doesn’t work. It’s the pilot that works, produces a good deck, and then sits there.
I’ve watched this happen often enough to stop treating it as bad luck. A pilot is designed to answer “can this work?” — and once it has, the project has achieved its stated goal. Nobody set the next goal, so nothing happens. The pilot becomes permanent. Eighteen months later someone commissions a second pilot of the same thing, because the first one’s results are now stale.
Why it’s structural
A pilot is cheap, low-risk and owned by whoever is curious. Production is expensive, carries real risk, and needs an owner with budget and a reason to spend it. Those are different people. The handoff between them is the actual project, and almost nobody plans it.
The tell is in how the pilot gets scoped. If the success criterion is a model metric — accuracy, MAPE, percentage improvement over baseline — there is no person on the other side waiting for the result. Real production projects get scoped against an operational number someone is already accountable for.
A better question to open with
Not “what should we pilot?” but “if this works, whose budget does it come out of next year, and what number on their scorecard moves?”
If nobody can answer that, you’re about to build something that will work perfectly and go nowhere. That’s worth knowing in week one rather than month nine.