What characteristics make a KPI useful for decision-making?
Most KPI failures are design failures, not measurement failures. A KPI earns its place only if a specific person can look at it, know whether it is good or bad, and know what to do next.
Unambiguously defined — one written formula covering inclusions, exclusions, grain and time treatment, so two analysts computing it independently get the same number.
Reliably measurable — sourced from a system of record with acceptable completeness and latency; a KPI whose underlying data is 20% missing will be argued about rather than acted on.
Sensitive and controllable — it responds within a useful horizon to levers the owning team actually holds; a metric nobody can influence generates anxiety, not decisions.
Comparable — it comes with a target, prior period or benchmark, since a bare number cannot be judged and a KPI without a comparison is just a reading.
Hard to game and paired — single metrics get optimised destructively, so pair a volume KPI with a quality one (tickets closed with reopen rate, sales with refund rate) to keep the incentive honest.
Key terms: metric definition, system of record, data completeness, latency, actionability, target, benchmark, guardrail metric, Goodhart's law
Do not recite SMART — interviewers hear it constantly and it signals a textbook answer. Instead give the characteristics with a failure mode attached to each: an ambiguous definition causes meetings about whose number is right, an uncontrollable KPI causes learned helplessness, an unpaired KPI gets gamed.
The guardrail-metric point is the strongest differentiator because it shows you have seen incentives distort behaviour; naming Goodhart's law is a bonus if you can also give the concrete example. Expect the follow-up "here is a KPI, critique it" — practise attacking one such as "number of reports produced", which is measurable but has no link to any outcome.