The Kill Rate is the share of AI use cases an organisation deliberately stops, tracked over time like delivery throughput. It sounds negative and is the opposite: because only about a third of believed-useful ideas ever move the metric, a portfolio with a healthy kill rate is one where judgement is actually operating and escalation of commitment is being beaten. A near-zero kill rate is the warning sign — it means continuation has become the default and zombie initiatives are accumulating.
It pairs with Default Off — the philosophy that production is earned, not the automatic destination of every promising experiment, so every transition (experiment → pilot → production → scale) is a fresh, evidence-gated decision rather than an automatic promotion. Hence the operational test: a pilot without a kill criterion is not a pilot, it is a delayed programme. The Kill Rate sits alongside the conviction-gap (whether to build at all) and contrasts with the production-gap (the worthwhile pilot that could not scale, as opposed to the one that should not). The depth treatment is Why Most AI Use Cases Should Never Reach Production.