From Product Theory: The Hidden Forces That Shape User Behavior — 40+ short chapters on why users behave the way they do.
A Soviet nail factory was graded on how many nails it produced. So it produced millions of tiny, useless nails. The quota was met. The point was missed.
When the metric switched to weight of nails produced, the factory started making enormous, impractical ones. Target hit again. Goal missed again.
That's Goodhart's Law: when a measure becomes a target, it ceases to be a good measure. And it isn't a story about central planning gone wrong. It's happening in your product right now, and the people gaming your metrics aren't villains. They're your own team, doing exactly what you asked.
Your metrics get gamed — by you
You set a goal: increase daily active users. Your team optimizes. They add notification spam. A few dark patterns. Some engagement tricks that boost the login count.
DAU rises. But are users actually getting value? Or did you just juice a number while quietly eroding the trust that made the product worth opening in the first place?
You didn't mean to game the metric. Nobody gathered in a room and decided to trick users. But the moment DAU became the target, people found the cheapest path to move it — and that path had almost nothing to do with the outcome you cared about.
The famous cautionary tales
The biggest platforms on earth learned this the hard way, at scale, in public:
- Facebook optimized for "time spent." The algorithm discovered that outrage keeps people scrolling. Time spent went up. User wellbeing went down. The metric was hit. The goal was missed.
- YouTube optimized for "watch time." The recommendation engine learned that increasingly extreme content — conspiracies, inflammatory takes — kept people glued to the screen. Watch time soared. So did the misinformation problem.
- Your support team gets measured on "tickets closed per day." They start closing tickets before issues are resolved, or split one simple problem into three tickets to inflate the count. The dashboard looks great. Customer satisfaction quietly tanks.
None of these teams were incompetent. They were obedient. That's the trap. Goodhart's Law punishes exactly the teams that take their goals seriously.
Why it happens: the mechanism
Charles Goodhart first noticed this in 1975, studying monetary policy. He observed that statistical relationships break down the moment you use them to steer. Marilyn Strathern later sharpened it into the version we quote today: any metric that becomes a target will be gamed.
The logic is almost embarrassingly simple.
A metric is a proxy. It represents something we actually care about — but it is not that thing itself.
The measure was useful because it correlated with the goal. When you optimize directly for the proxy, you sever the link between the proxy and the goal. People find ways to move the number without moving the outcome. And the tighter you squeeze, the faster that correlation dies.
Rule of thumb: The metric you target is the metric that stops working.
It's everywhere once you see it
This isn't a product-management quirk. It's a law of any system with a scoreboard.
| What we care about | What we measure | What we get |
|---|---|---|
| Learning | Test scores | Teaching to the test, grade inflation, cheating |
| Patient health | Wait times | Shorter recorded waits, not faster care |
| Content quality | Engagement | Clickbait, outrage, addiction |
| Sustainable business | Growth | Cash-burning vanity metrics |
Look closely and you'll notice the same pattern each time: the number went up, the thing it was supposed to represent went sideways or down.
When leading indicators start lying
The best metrics are leading indicators — they predict future success before it arrives. But the moment you target one, it stops predicting.
NPS is the classic casualty. Net Promoter Score was supposed to forecast growth. Then companies started gaming it: begging customers for 10s, cherry-picking who got surveyed, tying employee bonuses to the score. NPS inflated everywhere. Its correlation to actual growth quietly weakened. The instrument was calibrated to reality until we started leaning on it — then it detached.
What to do instead
The answer is not to throw out metrics. It's to stop treating them like truth and start treating them like tools.
1. Measure outcomes, not proxies — when you can
The closer a metric sits to the actual goal, the harder it is to game.
- "Revenue" is harder to game than "leads."
- "Retention" is harder to game than "signups."
- "Customer success" is harder to game than "tickets resolved."
Push your metrics toward outcomes. Accept the trade-off: real outcomes tend to lag and are harder to read in real time. So you'll still lean on leading indicators — just hold them loosely, and never mistake them for the finish line.
2. Use multiple metrics in tension
A single metric becomes a target. Multiple metrics pulling against each other become a system.
Measure only growth, and you'll sacrifice profitability. Measure only profitability, and you'll strangle growth. Measure both, and each one polices the other's cheating.
Frame goals as constraints, not solo objectives:
- "Increase DAU — but not at the expense of 30-day retention."
- "Reduce churn — but not at the expense of acquisition."
When a team games one number, the tension partner usually snaps and exposes it.
3. Rotate metrics periodically
A metric that never changes will eventually be perfectly gamed. If each quarter shifts the focus slightly, teams can't over-optimize any single number — they have to improve the underlying system to keep winning.
Yes, this is annoying for planning. It's also honest about what metrics can and can't do.
4. Watch for surrogation
Surrogation is when the metric becomes the goal in people's minds — when they forget it was ever a stand-in.
"We need to improve DAU" is surrogation. "We need more people getting daily value" is the actual goal. The words matter, because the language you repeat in standups is the goal your team will actually chase.
Keep the real goal visible. Metrics are servants, not masters.
The limit of the limit
None of this means metrics are useless. They're excellent for diagnosis. "DAU dropped 20% last week — what happened?" is a genuinely valuable question. Metrics are how you spot problems and ask better ones.
The danger is narrower than "measuring things is bad." The danger is when a metric becomes the sole definition of success — when hitting the number becomes the point, instead of a signal that you're achieving the outcome underneath it.
The nail factory didn't fail because it counted nails. It failed because counting nails became the entire job. Somewhere along the way, everyone forgot the factory existed to build things people could actually use.
Metrics serve the goal. They are not the goal. The day you forget that is the day your best number starts lying to you with a straight face.
If you find yourself squinting at your own dashboard tomorrow, wondering which number has quietly started lying to you, that's the idea doing its work. Subscribe to Context Limit and I'll send the next one along when it's ready.