MModern Wisdom
← All frameworks
StrategyGurwinder Bhogal

Goodhart's Law

When a measure becomes a target, it stops being a good measure.

Difficulty
Easy
Time to result
~ongoing to results
Steps
5
Confidence
82%

Goodhart's Law states that once a measure is turned into a target, people optimise the measure rather than the thing it was meant to track, so it ceases to be a good measure. The mechanism is simple: a metric is a proxy for a goal; humans reliably find loopholes; and effort flows to whatever is being counted. Gurwinder's example is British colonists paying for dead snakes in India, which caused people to breed snakes for the bounty. The same pattern shows up when follower counts proxy credibility (the untrustworthy game their way to the top) or when email-capture numbers replace genuine audience trust. The practical response is not to abandon measurement but to name the true outcome, anticipate how any single number can be gamed, and steer with multiple diverse signals while closing loopholes as they appear.

Origin

Named after economist Charles Goodhart. Gurwinder Bhogal popularised it in a viral Twitter thread of psychological mental models, discussed here with Chris Williamson using the cobra/snake-bounty and follower-count examples.

Core principles

  • 01Any single metric used to steer a system will eventually be gamed.
  • 02Humans reliably hunt for the loophole between the metric and the real goal.
  • 03The metric is a proxy; optimising the proxy can destroy the underlying goal.
  • 04System design is a perpetual cat-and-mouse game between rule-makers and gamers.

How to run it

  1. 1

    Name the real outcome

    Write down the actual goal you care about, separate from any number that stands in for it. The number is a proxy, not the prize.

    Pro tip If you cannot articulate the outcome without citing the metric, you have already collapsed the two.

  2. 2

    Red-team the metric

    Ask how a clever, self-interested person could maximise the number while completely failing the real goal.

    Watch out Assume people WILL find the loophole; history shows they always do.

  3. 3

    Steer with multiple diverse signals

    Replace one dominant metric with several that are hard to game simultaneously, so gaming one is exposed by the others.

  4. 4

    Watch for decoupling

    Monitor whether the metric is drifting away from the outcome over time, which signals it has become a target and is being gamed.

  5. 5

    Patch loopholes, expect new ones

    Close exploits as they surface, treating rule-setting as an ongoing game rather than a solved problem.

    Pro tip This is why laws accrete clauses: each new rule closes a freshly discovered loophole.

In the wild

The Indian snake bounty

British colonists in India wanted to reduce the snake population, so they measured progress by the number of snake corpses turned in and paid a bounty for each. People responded rationally to the incentive by breeding snakes purely to kill them for the reward. When the scheme was scrapped, the now-worthless bred snakes were released, leaving more snakes than before. The metric (dead snakes) had become the target and completely detached from the goal (fewer snakes).

The bounty increased the snake population instead of reducing it.

Follower counts as credibility

If you measure a person's trustworthiness by their follower count, people game it with underhand tactics to inflate followers. The result inverts the intended signal: the least trustworthy accounts often accumulate the largest followings, so the metric now points the opposite way to the goal it was meant to serve.

Follower count becomes an unreliable, sometimes inverted, proxy for credibility.

Common mistakes

Trusting a single headline number

Compressing a complex goal into one metric hands people a clear target to game. Diversify the signals you steer by.

Blaming the gamers, not the design

Gaming is the predictable human response to any exploitable rule. Fix the incentive structure rather than moralising about the behaviour.

Is it for you?

Best for

Managers, product owners, and policymakers who set targets or incentives that people will respond to.

Not ideal for

One-off decisions with no ongoing incentive structure or measurement to game.

From the transcript

When a measure becomes a goal, it ceases to become a measure.

Chris Williamson

history has shown that human beings will always game a system if they can do so

Gurwinder Bhogal

From the episode

15 Mental Models To Understand Psychology - Gurwinder Bhogal - #385

Gurwinder Bhogal