The OEC is the answer, written down before launch, to “what number decides this experiment?” Kohavi and colleagues borrowed the term from classical design of experiments and made it the centerpiece of online testing culture: a single criterion forces trade-offs to be negotiated once, aligns teams behind one objective, and blocks the temptation to pick a flattering metric after seeing the results.
The hard part is the tension baked into the definition: the OEC must be measurable within an experiment’s few weeks, yet predictive of long-term value. Raw short-term revenue fails the test (a worse search engine can raise queries and ad clicks short-term), which is why practical OECs lean on validated proxies like sessions per user, engagement thresholds, or trial conversion, with guardrail metrics standing watch over everything the OEC ignores.