Goodhart's Law says when a measure becomes a target it stops being a good measure. Here is why small teams accidentally reward the number instead of the outcome, and how to build metrics that resist gaming.
Wells Fargo paid a combined 185 million dollars in 2016 after employees chasing a cross-sell target opened more than two million accounts customers never asked for. That is Goodhart's Law in one sentence: when a measure becomes a target, it stops being a good measure. The number you reward is the number people will produce, whether or not it reflects the outcome you wanted.
I study behavioral psychology, and I find this pattern fascinating because it is so consistent. Set a target, attach it to pay or praise or job security, and people optimize for the target. Not out of malice. Because that is what the incentive told them to do.
Small teams are especially exposed. When five people share a dashboard, one gamed metric can quietly reshape how everyone spends their week.
The economist Charles Goodhart wrote the original version in 1975: "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes." The memorable phrasing came later, from anthropologist Marilyn Strathern in 1997.
Her line is the one worth taping to a wall: when a measure becomes a target, it ceases to be a good measure. A medical education editorial on the same principle puts it plainly. Once a metric is targeted for improvement, it loses its validity as a reliable indicator of the thing it was standing in for.
The psychologist Donald Campbell reached a parallel conclusion in 1976. Campbell's law holds that the more any quantitative indicator is used for decision-making, the more it will be corrupted, and the more it will distort the process it was meant to monitor. Two thinkers, two fields, one warning.
Gaming rarely looks like cheating. It looks like a reasonable person responding to the scoreboard in front of them.
The oldest illustration is a bounty. During the 1902 rat bounty in Hanoi, officials paid for severed rat tails to shrink the population. Catchers responded by clipping tails and releasing the rats to breed, and some began farming rats outright. The target was tails. The goal was fewer rats. The town got more.
That gap between target and goal is the whole story, and it is why the pattern is often called a perverse incentive. People are not failing to hit the number. They are hitting it too well, in the cheapest way available, which is usually not the way you imagined.
This connects to something I have written about in how invisible bias steers technology decisions. The bias here is that a clean number feels like truth. It feels objective. So we trust it long after the behavior underneath it has bent to fit the measurement.
Some proxies are almost designed to be gamed. A few show up in nearly every small business I talk to.
Support tickets closed is a classic. Reward volume and you reward fast closes, reopened tickets, and problems that were never actually solved. The healthier goal, problems resolved so they do not return, is harder to count, so it loses to the easy proxy.
Calls made or emails sent is another. When activity is the target, a salesperson can have a spectacular week by the dashboard and a terrible one by the pipeline. This is why I keep pointing teams toward the metrics that actually connect to revenue rather than the ones that are simply easy to log.
Software teams see it with velocity. Story points were built as an estimation aid for planning, not a productivity score. Turn them into a target and estimates inflate on their own, which is one reason Martin Fowler argues that measuring developer productivity by output is a fool's errand. The real measure is business value delivered, and that resists a tidy weekly number.
You cannot delete the effect. You can design around it. Three practices do most of the work.
First, pair every metric with a counter-metric. Speed pairs with quality. Tickets closed pairs with reopen rate. Calls made pairs with conversion. A pair is much harder to game than a single number, because winning on one while wrecking the other becomes visible immediately.
Second, measure the outcome, not the proxy, wherever you can afford to. Instead of tickets closed, track whether the customer came back with the same issue. Instead of accounts opened, track accounts actually used. What client retention really measures is a better signal than almost any activity count, because a customer staying is the outcome, not a stand-in for it.
Third, watch what gets gamed and treat it as information. If a number jumps while customers do not feel a difference, the metric is telling you where the pressure is landing. Wells Fargo had that signal for years. The bank quietly fired employees for opening fake accounts long before it fixed the target that produced them.
None of this requires a bigger analytics stack. It usually requires connecting the data you already have, which is its own quiet problem, because disconnected data is more expensive than missing data when you are trying to see a proxy and its outcome side by side.
There is a human cost to targets that ignore this law. When a metric is unreachable by honest means, people do not usually quit. They cut corners, hide the strain, and keep hitting the number until something breaks.
That is one of the ways burnout hides in plain sight. A team gaming an impossible target can look like your highest performers right up until they leave. The dashboard stays green while the people behind it run out.
The calm version of this is not to abandon measurement. It is to hold every number loosely, knowing it is a proxy, and to keep asking what outcome it was supposed to represent. At Kief Studio, we build reporting the same way we build systems, so the measure and the outcome stay close enough that gaming one shows up against the other. Good instrumentation is a byproduct of good engineering, not a scoreboard you point at your team.
It also helps to remember why the wrong number felt so trustworthy in the first place. That is the same territory as the sunk cost fallacy running a technology stack: we defend a measure because we have invested in it, not because it still tells us the truth.
Goodhart's Law says that when a measure becomes a target, it stops being a good measure. The moment you reward a number, people optimize for that number, and it drifts away from the outcome it was supposed to represent.
No. Measurement is how you learn what is working. The risk is treating a single proxy as the goal itself. Pair each metric with a counter-metric and measure real outcomes where you can, and numbers stay useful.
Watch for a number that climbs while customers, revenue, or quality stay flat. That divergence is the tell. It usually means people have found a cheap way to hit the target that does not move the outcome underneath it.
They describe the same effect from different fields. Goodhart's Law comes from economics and focuses on targets collapsing under pressure. Campbell's law comes from social science and stresses how using an indicator for decisions corrupts the process it measures.
Measure outcomes, not proxies, and never let a single number stand alone. Speed with quality, volume with retention, activity with results. Balanced pairs are far harder to game than any figure on its own.
Martin Seligman's research shows resilience isn't a personality trait — it's a skill. Frontiers in Psychology's 2025 study found that founders with higher psychological capital have measurably lower burnout. This isn't motivational content. It's operational infrastructure.
I have credentials in marketing psychology, consumer behavior, and positive psychology. Not because I wanted to be a therapist — because I wanted to understand why people do what they do.
Google Analytics tells you what happened. Behavioral analytics tells you why. The difference between the two is the difference between data and insight.
Work With Us
Kief Studio builds, protects, automates, and supports full-stack systems for businesses up to $50M ARR.
Newsletter
Strategy, psychology, AI adoption, and the patterns that actually compound. No spam, easy to leave.
Subscribe
