Walk into most mid-size enterprises and you'll find three to a dozen AI pilots in flight — a chatbot here, a forecasting notebook there, an agent demo somebody built over a hackathon weekend. Walk in a year later and most of the same pilots are still pilots: technically working, organizationally orphaned, never converted into a funded, owned, production line item. The pattern has a name worth naming: pilot purgatory. It isn't a technology failure. It's a governance one.
Why pilots stall instead of graduate
- No named business owner — the pilot was sponsored by IT or an innovation team, not by the person whose KPI it would actually move. Nobody with budget authority feels ownership.
- Success was never defined as a number — 'this looks promising' is not a graduation criterion. Without a target, there's no moment where the pilot has 'succeeded' enough to fund.
- The cost model was never built — a pilot that ran on a free tier or a grant-funded sandbox has no answer for 'what does this cost at ten times the users,' which makes the budget conversation impossible to start.
- It solved a problem nobody urgent enough owned — technically interesting, organizationally optional. Optional initiatives lose every prioritization meeting to urgent ones.
Three gates a pilot must pass to earn a production budget
- Technical proof on real data — not a curated demo dataset, but your actual messy production data, with an honest accuracy number attached.
- A named owner who changes their workflow — a specific person or team who will use the system's output to do their job differently, and who's on record saying so.
- A cost model that doesn't scale linearly with usage — if serving ten times the users costs ten times as much with no efficiency curve, the pilot isn't a product, it's a subsidy.
The unglamorous fix: kill pilots faster, not slower. A pilot that fails one of the three gates in month two should be closed in month two, not kept alive as a slide in a quarterly review. The budget it frees funds the pilot that actually clears all three.
What this means for the initiatives you fund next
Before greenlighting a new pilot, write down what would make you kill it and who owns the decision to graduate it. If you can't answer both, you're not funding a pilot — you're funding an indefinite experiment, and those are the ones still running, unowned, a year from now.