A good one measures delivered value rather than activity, and revenue tends to follow it rather than being it. Signups, pageviews and downloads fail the test — they can all rise while the product is getting worse. The metric should be something that only goes up when a customer genuinely got what they came for, and it should be countable weekly without a research project.
The point is alignment and refusal. With one number agreed, arguments about what to build next become empirical: does this plausibly move it. Without one, every feature request sounds equally reasonable and roadmaps get decided by whoever asked most recently. For a solo founder the value is the same in miniature — it converts a hundred possible tasks into a filter, which is exactly what a limited week needs. Pair it with one or two guardrail metrics, such as churn or gross margin, so you do not optimise the headline number into a worse business.
A concrete example: a document tool could track signups, but signups tell you nothing about value. "Documents completed and shared per week" only rises when someone finished real work. Suddenly the priorities reorder themselves: a faster editor moves it, a redesigned marketing page does not, and an onboarding flow that gets a first document finished on day one moves it most of all. Same team, same week, better choice.