Goodhart's Law & Metric Gaming

Goodhart's Law states that when a measure becomes a target, it stops being a useful measure. This happens because people start to 'game' the metric—finding the easiest way to hit the number, even if it undermines the original purpose. For instance, if customer support agents are rewarded solely for 'calls handled per hour,' they might rush customers off the phone to improve their stats, leading to worse actual service.

Ask this: What is the real goal here, and is this metric just a proxy for it?

The core of Goodhart's Law is that people respond to incentives. When we elevate a proxy metric (like 'calls handled') to a high-stakes target, we incentivize people to optimize the proxy itself, not the ultimate goal (like 'happy, loyal customers'). This is known as metric gaming or surrogate optimization.

This pattern appears in many different domains. In schools, when standardized test scores become the sole measure of success, educators may be pressured to 'teach to the test,' focusing on test-taking strategies rather than fostering deep, critical thinking. Student scores might go up, but their overall education could suffer.

In the corporate world, a sales team judged only on 'number of new accounts' might sign up low-quality clients who cancel after a month, hitting their quota but failing to build sustainable revenue for the company. Similarly, a social media platform aiming to maximize 'time on site' might promote polarizing or addictive content, which boosts the metric but harms user well-being.

The key isn't to abandon measurement. Instead, it's to be smarter about it. Effective strategies include using a 'basket' of counter-balancing metrics (e.g., measuring both speed and quality), incorporating qualitative feedback, and regularly questioning whether your metrics are still truly aligned with your goals.

Originally observed in economics, Goodhart's Law has become a widely cited heuristic in management, public policy, and data science. It describes how the act of measuring and targeting a social or economic indicator inevitably alters the behavior of those being measured, distorting the indicator's reliability.

• Goodhart (1975): First articulated the core idea in the context of monetary policy, stating, "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes." [1]
• Strathern (1997): Popularized the most common phrasing of the law and extended its application to the culture of auditing and performance management in academia, highlighting how accountability systems can create perverse incentives. [2]
• Muller (2018): In The Tyranny of Metrics, he documents dozens of cases of 'metric fixation' and its negative consequences, arguing for more judicious use of quantitative measures and warning against using metrics designed for diagnosis as tools for reward and punishment. [3]

Goodhart's Law is a heuristic, not an iron law of nature. Its effects vary based on context. Some metrics are inherently more resistant to gaming than others; for example, net profit is harder to manipulate than 'lines of code written.' The strength of the effect also depends on the stakes involved—high-stakes targets tied to bonuses or promotions are much more likely to be gamed than low-stakes diagnostic indicators. The law serves as a crucial warning about the design of incentive systems, not a rejection of measurement itself.