When a measure becomes a target, it stops being useful

A number that describes something well is tempting to manage. If satisfied customers write more reviews, raise the number of reviews. If good developers close more tickets, count closed tickets. The reasoning feels sound, and it fails in a specific, repeatable way.
Charles Goodhart made the observation in 1975, as an aside in a paper about British monetary policy. The relationship between money supply and the wider economy had been stable enough to look like a control lever, and his warning was that using it as a lever would destroy the stability that made it attractive.
"Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes."
The version most people know today came from the anthropologist Marilyn Strathern, writing in 1997 about audit culture in British universities: when a measure becomes a target, it ceases to be a good measure. Donald Campbell had described the same dynamic in education even earlier, noting that achievement tests are informative indicators of school performance right up to the moment they become the goal of teaching.
The reason is that we almost never measure what we care about. We measure a proxy. Nobody can measure understanding, so we measure test scores. Nobody can measure engineering value, so we count merged pull requests. The proxy correlates with the goal as long as people are pursuing the goal. Point an incentive at the proxy and you have asked for the correlation to be broken, because the cheapest way to move a proxy is rarely to improve the thing it stood for.
The result is not usually fraud. It is a quiet redistribution of effort. Tickets get split so more of them can be closed. Difficult cases get avoided because they damage a success rate. Response time improves because replies are fast and empty. Everyone is doing what they were asked to do, and the metric is the only part of the system that still looks healthy.
Which does not mean measurement should be abandoned. Working without numbers means trusting impressions, seniority and volume of voice, which is worse. It does mean treating metrics with more suspicion than we usually do.
A few things help. Keep some measures purely for observation and never attach consequences to them, so at least part of your instrumentation stays honest. Pair any quantity measure with a quality or cost measure that moves in the opposite direction when the first one is gamed. Ask, before publishing a target, what the cheapest way to hit it would be, and assume someone will find it. And when a number improves sharply while nothing around it feels different, treat that as a finding rather than a success.
A measure is a description of reality. A target is an instruction. The moment we confuse the two, the number keeps rising and stops telling us anything.
- Charles Goodhart — “Problems of Monetary Management: The UK Experience”
- Marilyn Strathern — “‘Improving Ratings’: Audit in the British University System”
- Jeffery Rodamar — “There Ought to Be a Law! Campbell versus Goodhart”, Significance (2018)
- Jerry Z. Muller — The Tyranny of Metrics (2018)
— Charles Goodhart