Goodhart’s law
Numbers are how large organisations see. But a number is only a reading of the thing you care about, and the moment people are rewarded for the reading, many will find ways to move it that leave the thing itself untouched — or make it worse.
The idea
The economist Charles Goodhart made the observation in 1975 about monetary statistics: a measured relationship tends to break down once it is used as a target for control. The anthropologist Marilyn Strathern later put it in the form most people quote: “When a measure becomes a target, it ceases to be a good measure.”
The mechanism is simple. Almost any measure is correlated with the goal for a reason: under ordinary conditions, the easiest way to move the number is to do the real work. Put pressure on the number and people look for cheaper ways to move it. Those ways exist for nearly every measure, and pressure finds them.
The call that ended too soon
Call centres have long tracked average handle time: the minutes an agent spends per call. It is a sensible number to watch. A centre where calls are getting longer may have a confusing product, a missing help page or agents who need training. The trouble starts when agents are ranked or paid by it. Calls can be shortened by solving problems faster — or by rushing, transferring the caller elsewhere, or ending the call before the problem is fixed. The dashboard improves either way.
A simple model shows how far apart the two can drift. Say a careful call takes 8 minutes and solves the problem 90% of the time, while a rushed call takes 3 minutes and solves it a quarter of the time. As pressure grows, a larger share of calls is rushed, and every unsolved problem calls back:
This is why many service teams pair call length with a second number that moves the other way when someone games the first: the share of problems solved on the first call, or repeat calls within a week.
Software teams meet the same pattern with test coverage, the share of code that is run by automated tests. Coverage is a useful warning light: code that no test runs has certainly not been tested. But a team told to reach 90% can write tests that run the code and check nothing. The number goes up; the protection does not.
Try it yourselfWhat happens when the number becomes the target?Put pressure on call length and compare the dashboard with what callers get.How to use it
- For every number you track, write down the goal it stands for, in one sentence.
- Ask how someone could move the number without moving the goal. If you can think of a way, so can they.
- Pair each target with a counter-measure that the cheap shortcut would make worse.
- Keep some measures for learning only, and do not tie pay or ranking to them.
- Look at real cases regularly — listen to calls, read tests — not only at the totals.
Where it misleads
Goodhart’s law is not an argument against measuring. Without numbers, problems are found late and arguments are settled by whoever speaks loudest. And some measures are hard to game because they are close to the goal itself: the share of payments that go through is both the measure and much of the point.
It also does not mean that people are dishonest. Most gaming is not deliberate; it is the slow drift of attention towards whatever is being counted and praised.
In one line: measure what matters, but do not pay for the reading alone.