Pilot success criteria (exit criteria)
Measurable conditions agreed before a pilot starts that decide whether it passed and the buyer moves to a full contract.
Also called exit criteria or, when both sides sign them, a mutual success plan. Good criteria name a metric, a baseline, a threshold and a test set: for an AI assistant, for example, the share of answers a reviewer accepts on a fixed set of real questions. Without them a pilot ends in opinions, and opinions rarely unlock a budget.
Related terms
- Proof of concept vs pilot (PoC, POV): A proof of concept shows something can work on sample data; a pilot runs it for real users on real data; a proof of value measures whether it pays.
- Pilot purgatory (PoC purgatory): When AI proofs of concept and pilots keep running or get dropped without reaching production, or go live but cost more in human oversight than they save.
- Golden dataset: A fixed set of real inputs with expected answers or checks, used to test an AI feature after every prompt or model change.
- Definition of Done: A shared checklist of what must be true before work counts as finished: tested, reviewed, deployed, documented.
