What a pilot is allowed to skip that production cannot
Pilots run on clean, hand-picked data, with a human quietly fixing edge cases behind the scenes, and no real accountability if it's wrong for a week. None of that survives contact with production, where the data is messy by default and an error has a named owner.
A pilot plan that doesn't specify who owns production data quality and who's accountable for a wrong output isn't a pilot — it's a demo with a future tense attached.
The handoff that most organisations never plan for
Pilots are usually owned by an innovation function; production ownership belongs to whoever runs the system of record it touches. If that handoff — budget, on-call, data contract, and all — isn't designed before the pilot starts, it becomes the reason the pilot quietly stalls at the finish line.
A successful pilot and a production-ready system are judged by completely different criteria. Confusing the two is the most common reason good pilots die.MNB AI Practice
Key actions
- Name the production data owner before the pilot starts, not after
- Define the accountability chain for a wrong output in advance
- Plan the handoff from the innovation team to the owning team explicitly
- Measure the pilot against production-realistic data, not a curated set