When a Metric Becomes a Target, It Stops Being a Good Metric

Goodhart's Law: when a measure becomes a target, it stops being a good measure. The moment a number decides someone's bonus, promotion or reputation, people find the shortest path to the number — and the shortest path is almost never the thing the number was chosen to represent.

This is not cynicism about people. It is a structural fact about proxies. Any metric is a simplification of something you actually care about, and simplifications have gaps. Pressure finds the gaps.

What it looks like in the wild

• Schools judged by test scores teach to the test. The score rises; general literacy does not.
• Support teams measured on tickets closed close fast and reopen often. Handle time drops; the same customer contacts you four times.
• Sales measured on calls made make short calls. Activity rises; revenue does not.
• Developers measured by lines of code or story points write verbose code and inflate estimates. Velocity charts look great; the product does not ship faster.
• Publishers measured on pageviews split one article into eight pages. Views double; readers leave.
• Hospitals given waiting-time targets learn to record the clock differently. The dashboard turns green.

In every case nobody had to be dishonest. They were told what success looks like, and they delivered exactly that.

Why it happens so reliably

1. The proxy is easier to move than the goal. "Be a better teacher" is hard and slow. "Raise the score on this test" is specific and has known techniques.
2. Measurement changes behaviour by design. That is why you measure. You cannot have the motivating power without the distorting power.
3. The gap is invisible from above. The dashboard shows the proxy, not the goal, so improvement and gaming look identical from the top.

Four ways to keep a number honest

Pair every metric with a guardrail that moves the other way. Speed with reopen rate. Volume with quality sampling. Growth with churn. A single number can be gamed; a pair that must both hold is much harder to fake.

Treat metrics as thermometers, not steering wheels. A thermometer tells you to investigate. A steering wheel tells everyone what to optimise. The same number in a weekly review means something completely different from the same number in a compensation formula.

Run the pre-mortem question. Before publishing a target, ask out loud: "How would I hit this number without doing the real work?" Everyone in the room usually knows the answer in ten seconds. Whatever they say is what you will get. If you cannot close that path, do not use that metric as a target. This is the same reflex as imagining the project already failed and asking why.

Keep a qualitative channel open. Read ten support conversations a month. Watch one user try the product. Sit in one class. The number tells you that something moved; only direct contact tells you what actually happened.

What this does not mean

It does not mean stop measuring. Unmeasured work drifts toward whoever is loudest, and "we just know" is a worse instrument than an imperfect number. The rule is narrower and more useful: measure honestly, target carefully, and never tie hard consequences to a single proxy. Retire metrics when they stop being informative; a metric that has been a target for two years is usually telling you about the gaming, not the goal.

Frequent questions

Is this the same as Campbell's Law? Nearly. Campbell's version is about social indicators used for decision-making becoming corrupted; Goodhart's came from monetary policy. Same mechanism, different field.

My manager demands one KPI. What do I do? Accept the KPI and volunteer the guardrail: "Happy to own this number — let's also watch this second one, so we know the first is real." Framed as protecting the result, not resisting accountability, this is rarely refused.

How do I choose a metric that resists gaming? Prefer outcomes to activity, prefer things a customer would notice, and prefer numbers that are expensive to fake. Repeat purchases resist gaming better than satisfaction surveys; a working demo resists gaming better than a progress percentage — see measuring progress with small demonstrations.

How do I predict where a team will drift? Look at what they are measured and paid on, not at what the mission statement says. Follow the incentives and you will predict the behaviour — and if you want influence inside an organisation, start by understanding what other people are measured on.