Goodhart's Law
The moment a measure decides rewards, it starts changing the work.
Once people are rewarded for a number, they can improve the number while the thing it was meant to show gets worse.
A metric can change the work without anyone cheating
We use measures because we need a way to see whether something is improving. Trouble starts when the measure also decides status, bonuses, promotions or blame. People then have a reason to organise around the number, even when that pulls them away from the outcome the number was meant to stand for.
Nobody has to lie. They can take the easier work that counts, avoid evidence that might hurt the target, or spend less time on work the measure cannot see.
That is Goodhart's Law: when a measure becomes a target, it stops being a reliable measure of the thing we cared about.
When 92% delivery hides a weaker team
An engineering team is judged by the share of roadmap items it delivers. Its delivery rate rises from 65% to 92%.
The number looks great. Behind it, teams choose more predictable work, split initiatives into smaller tickets, avoid foundational improvements and keep quiet when new evidence should change the roadmap. Incidents rise, while users see little improvement.
The target rewarded exactly this behaviour even though it did not represent the full job.
Where this shows up
Education. Test scores can tell us something about learning. When scores become the goal, teaching narrows to what can be examined, and curiosity or lasting understanding gets less attention.
Operations. A target for closing tickets quickly can teach a support agent to close the record before the customer's problem is solved. The closure rate improves, then the customer reopens the same issue and support does the work twice.
Product. Optimising clicks can make an interface more provocative. Clicks rise exactly as expected, while trust and long-term value fall because the number captures too little of the experience.
Personal life. A reading streak can turn reading into keeping a counter alive. Someone keeps the streak by choosing short books they do not want to read.
Design around the outcome
Before relying on a measure, ask: How could this number improve while the real outcome gets worse?
Then add a guardrail that would expose that failure, use direct evidence from the people affected and keep checking whether the measure still tracks the outcome. A number can help people pay attention. It should not get to define success by itself.