Charles Goodhart was writing about monetary policy in 1975. Central banks had noticed that certain measures of money supply tracked inflation reliably, so they began targeting those measures. The relationships promptly fell apart.

His conclusion generalizes further than he intended: any observed statistical regularity tends to collapse once pressure is placed upon it for control purposes. The version everyone quotes is Marilyn Strathern's tightening of it — when a measure becomes a target, it ceases to be a good measure.

Why it happens

A measure is a proxy. It correlated with the thing you care about because both were produced by the same underlying behavior, and nobody was trying to move the proxy directly.

Set it as a target and you change that. Now there are two routes to a higher number: do the underlying thing better, or affect the measure without doing the underlying thing. The second route is almost always cheaper, and people find it without anybody deciding to cheat.

Measure a sales team on calls made and you get calls made. Short ones, to people who will not buy, because a short call to an unqualified prospect raises the number at the lowest cost. Everyone involved is doing their job as defined.

The three shapes it takes

The proxy gets gamed. The number moves and the thing does not. Support measured on tickets closed produces closed tickets and reopened ones.

The proxy gets optimized at the cost of everything unmeasured. Nothing is gamed. Effort simply moves from what nobody counts to what somebody does. A team measured only on delivery speed writes worse code, and the speed number is honest.

The proxy becomes the goal. This is the slowest and worst. After a few years nobody remembers what the measure stood for, and hitting it is the job. That is what performance theater is — work shaped entirely around being seen to work, with the original purpose no longer in anyone's head.

What it does not mean

Stopping measuring is not the lesson, and it is the lesson most often taken.

An unmeasured business is worse at everything, and measurement is most of what makes a company sellable. The claim is narrower: a single measure used as a target will degrade. That is about using one number for control, not about knowing your numbers.

Working around it

Nobody has a fix. There are four things that help, and each one makes a target harder to hit without doing the actual work.

Pair every target with the thing it could destroy. Calls with conversion rate. Speed with defect rate. Revenue with retention. Nobody can game both directions of a pair at once, and they will try, which tells you something.

Measure outcomes rather than activity. Revenue retained resists gaming better than meetings held, because it is further from anything an individual can directly manipulate.

Change the measure before it is gamed, not after. A target that has been in place for three years is being managed to rather than being measured. Rotating what you watch keeps the correlation alive.

Keep the measure away from compensation where you can. Pay makes the pressure enormous and makes the law arrive faster. Watch a number to learn; pay on a number and you have bought whatever it counts.

The thing worth holding onto: when a metric improves sharply and the business does not feel different, the metric is being hit rather than earned. That gap is the signal, and it is usually visible a long time before anyone admits it.