Pattern/Mental Model/No. 0437
Goodhart’s Law
When a number becomes the goal, people find ways to move the number that don’t move the goal.
Also called When a measure becomes a target
- Evidence
- Well established
- Read
- 5 min
- Links
- 11 connections
01You've seen this when…
- at work
Support agents are measured on average call length. Calls get shorter. Customers call back three times.
- in life
You set yourself a goal of 10,000 steps a day. Some nights you find yourself pacing the kitchen at 11 p.m. Your step count goes up, but your fitness barely changes.
- out in the world
Schools judged on test scores spend more time on test practice and less on the subjects the tests were meant to sample. Scores rise faster than learning.
02The idea
A measure is useful because it moves along with something you care about. Call length loosely tracks how efficiently problems get solved. Test scores loosely track learning. But that relationship was observed while nobody was trying to move the number.
Once you reward the number, people and systems look for the cheapest ways to move it. The cheapest ways often leave the underlying thing untouched, so the link breaks. Goodhart, writing about monetary policy, put it as a statistical regularity that “will tend to collapse once pressure is placed upon it.” Marilyn Strathern’s later wording is the one people quote: when a measure becomes a target, it ceases to be a good measure.
This usually isn’t cheating. Mostly it’s honest people doing exactly what they’re rewarded for.
03Why it happens
The link between measure and goal can break in a few different ways:
- Narrowing. Effort flows to what’s measured and away from what isn’t. Teaching to the test is the classic case.
- Gaming. The recording changes while the reality stays put. Cases get reclassified, timing shifts, results bunch just over a threshold.
- Pushing past the range. The relationship held across normal values and breaks at the extremes. Shorter calls were better up to a point; past it, rushing callers off the line leaves their problems unsolved.
- The people measured know more. Those closest to the work see the shortcuts long before the people who set the target.
Pressure amplifies all of these. The bigger the bonus or the penalty, the more creative the optimization. Donald Campbell’s version, Campbell’s law, makes that point about public policy.
04A worked example
From the early 2010s, Wells Fargo pushed its branch staff to sell more products to each customer, with demanding daily sales goals. The metric was meant to reflect deeper customer relationships: someone with a checking account, a savings account and a credit card is a loyal customer.
What it looked like Cross-selling numbers that led the industry.
What was actually going on Under pressure, employees opened accounts that customers had never asked for. In 2016 regulators fined the bank $185 million. A later review put the number of potentially unauthorized accounts at around 3.5 million. The number went up while the thing it stood for, customer trust, went down.
What would have helped Pairing the sales metric with counter-metrics that are hard to fake at the same time, like accounts actually used and customer complaints. Auditing a sample of the numbers. And loosening the tie between pay and any single figure.
05How to spot it
06What to do about it
- Ask the cheapest-route question before you set a target. How could someone hit this without doing what we want? That’s inversion applied to metrics.
- Measure in pairs. Speed alongside quality, volume alongside returns. It’s harder to game two numbers that pull against each other.
- Keep some measures for learning only. A number nobody is paid on stays honest longer.
- Look at the thing itself. Every so often, talk to the customers, read the code, sit in the classroom.
- Rotate and audit. Change targets before people learn their shortcuts, and spot-check the numbers.
07Where it doesn’t bite
Goodhart’s law is a warning about how you use measures. Targets often do improve the thing itself, especially when the measure is the goal. If what you care about is literally how fast outages get fixed, measuring time to fix is fine. The law bites hardest when the measure is a loose proxy, the stakes are high, and the people being measured control how the number gets produced.
08Roots
Charles Goodhart was an economic adviser at the Bank of England in the 1970s, when British policy leaned on targets for the growth of the money supply. Relationships between money and the wider economy that had looked steady in the data started slipping once the Bank tried to steer by them. In a 1975 paper he summed it up as a half-joking law.
He wasn’t alone. In 1976 the economist Robert Lucas argued that relationships estimated from past data break down when policy changes, because people adapt to the new policy, an argument now called the Lucas critique. The same year, the social scientist Donald Campbell described how indicators used for decisions get corrupted, with examples from schools, policing and public programs.
The version people quote came two decades later and from another field. In 1997 the anthropologist Marilyn Strathern, writing about audits of British universities, gave it the short form that stuck. From there it travelled into management, healthcare and machine learning, where a system that optimizes a proxy at the expense of its goal is an everyday hazard.
09How solid is this?
Documented across monetary policy, management, education, healthcare and machine learning through case studies and natural experiments. How hard it bites depends on how loosely the measure tracks the goal and how much rides on it.
10Connections
- Often confused withCampbell's Law, Principal-Agent Problem
- Can lead toSpecification Gaming, Cobra Effect
- Part of Map vs. Territory
- See alsoSurrogate Endpoint Problem, Proxy Objective, Incentive Compatibility, Inversion, Leading vs. Lagging Indicator, Pareto Principle
+ 1 more in the list
11Origin and sources
Economist Charles Goodhart (1975), writing about monetary targets at the Bank of England. Anthropologist Marilyn Strathern’s 1997 phrasing made it famous. Donald Campbell described the same pattern in social policy (1976).
- [1]Goodhart, C. A. E. (1975). Problems of monetary management: The U.K. experience. Papers in Monetary Economics, Vol. I. Reserve Bank of Australia.
- [2]Strathern, M. (1997). 'Improving ratings': Audit in the British University system. European Review, 5(3), 305–321.
- [3]Lucas, R. E. (1976). Econometric policy evaluation: A critique. Carnegie-Rochester Conference Series on Public Policy, 1, 19–46.
- [4]Campbell, D. T. (1979). Assessing the impact of planned social change. Evaluation and Program Planning, 2(1), 67–90.
- [5]Manheim, D., & Garrabrant, S. (2018). Categorizing variants of Goodhart's Law. arXiv:1803.04585.
Suggest an edit· Updated 2026-10-01