KPIs That Corrupt

Key takeaways

• When a measure becomes a target, it tends to stop measuring the thing you actually cared about. Economists call this Goodhart's Law, and it is one of the most reliable forces in any organisation.
• People optimise the number, not the goal behind it. The cleverer and the more pressured they are, the faster the gap between the two opens up.
• History keeps running the same experiment: the 1696 window tax, McNamara's body counts in Vietnam, Wells Fargo's bonuses for opening accounts.
• The corruption is rarely fraud. It is usually rational people responding sensibly to the scoreboard you built for them.
• The repair is less about finding better metrics and more about watching the behaviour a metric produces, and being willing to retire a number once it starts driving the wrong thing.

A measure that becomes a target stops measuring what you wanted

In the previous tier of this series I argued that a KPI is a map, a useful simplification of the business. This post is about what happens to the map when you start paying people to move it. The short version has a name: Goodhart's Law, usually stated as "when a measure becomes a target, it ceases to be a good measure." The moment a number controls rewards, status, or survival, people stop treating it as a description and start treating it as the objective. They will make the number go up. Whether the thing underneath it goes up is a separate question, and the answer is often no.

England taxed windows in 1696 and got darker houses

The cleanest illustration is three centuries old. In 1696, England brought in a tax on windows, on the sensible-sounding logic that wealthier people had bigger houses with more windows, so windows were a decent proxy for wealth. As a way to measure wealth, it was reasonable. As a target, it was a disaster. People bricked up their windows to cut their bills. Houses grew darker and worse ventilated, and public health suffered for it. The proxy and the goal came apart the instant the proxy carried a price. Nobody in the story was being irrational. They were doing exactly what the rule rewarded.

McNamara counted what was easy and missed the war

Robert McNamara ran the United States defence department through much of the Vietnam War as a believer in measurement, and he reached for the numbers that were easy to count. Body counts. Sorties flown. Tonnage dropped. The difficulty is that the things you can count are not always the things that matter, and a war is full of decisive factors that do not fit in a spreadsheet: morale, legitimacy, whether the population is with you. By optimising the measurable, the effort drifted away from the actual objective. The lesson is sometimes called the McNamara fallacy, and it lives on in every business that manages what is easy to track and ignores what is hard, simply because the easy thing has a dashboard tile and the hard thing does not.

Wells Fargo paid for opened accounts and got millions of fake ones

The modern version is almost too neat. Wells Fargo set aggressive targets for the number of accounts and products its staff sold, and tied bonuses and job security to hitting them. Staff hit them. They opened around two million accounts customers had never asked for, some entirely fictional, because the bank had made "accounts opened" the thing that paid, and "accounts opened" turned out to be very easy to manufacture and only loosely connected to "customers served". You can write this off as a few bad apples. It is more honest to see a barrel built to grow exactly this fruit. As the line attributed to Upton Sinclair goes, it is hard to get a person to understand something when their salary depends on their not understanding it.

The people gaming the metric are usually being rational

It helps to drop the language of cheating. In most cases the staff bending a metric are not villains. They are responding, sensibly, to the incentives in front of them, the way water runs downhill. A support team paid on tickets closed will close tickets fast, including by marking hard problems resolved when they are not. A performance team judged on blended return will quietly cut the brand spend the return figure cannot see, even as the brand weakens beneath them. None of them woke up wanting to harm the business. Each made the locally rational move on the scoreboard they were handed. The old observation about envy sits underneath all of it: people watch what gets rewarded around them and adjust, relentlessly, whether or not anyone intends them to.

I watched a version of this with a brand that decided to reward its customer-service team on average response time. Within a month the response time looked superb. It turned out the team had learned to fire off a fast, near-useless first reply to stop the clock, then deal with the real problem later, or not at all. Customers were angrier than before and refunds rose, but the one number on the wall was the best it had ever been. The metric was not lying. It was measuring precisely what it had been told to measure, which was no longer the thing anyone cared about.

Watch the behaviour, not just the number

If Goodhart's Law cannot be repealed, it can at least be managed, though the work never ends. The most useful move I know is to watch the behaviour a metric produces, not only the metric itself. When response time drops, go and read the replies. When accounts opened jumps, ring some of the customers. Pair every efficiency number with a quality number that pulls in the opposite direction, so that gaming one shows up as damage in the other. And keep Karl Braun's discipline from the briefing post in mind: insist on the why behind a target, because a number with no living purpose is the easiest of all to corrupt.

The harder truth is that no metric survives strong enough incentives forever. There is no one true KPI that cannot be gamed; the job is to keep noticing when a measure has started to drive behaviour you did not intend, and to have the nerve to change or kill it before the gap becomes the culture. That nerve is rarer than it sounds. Retiring a number that executives have celebrated for two years means admitting it was steering you wrong, and organisations are not built to enjoy that admission.

A good deal of what we do at Mean Decisions ends up here: sitting with founders and teams to work out which of their numbers are quietly corrupting the behaviour around them, and which are still telling the truth. It is less about adding metrics and more about asking, of each one, what is this actually making people do. The part I cannot resolve in an essay is that the answer keeps changing, because the moment people learn how they are measured, they begin, gently and rationally, to respond.