Goodhart’s Law in the Workplace

A support team tracks how quickly it answers customer questions. At first, response time helps managers spot backlogs. Once bonuses depend on speed alone, workers may send quick but incomplete replies, and the number no longer represents good support.

If a clinic rewards appointments kept strictly on schedule, staff may feel pressure to rush conversations that need extra attention. The schedule remains useful, but it needs to be considered alongside care quality and the complexity of each visit.

This is the workplace form of Goodhart’s Law: when a measure becomes a target, it can stop being a good measure. The phrase is associated with economist Charles Goodhart’s observation about policy indicators, and it captures a broader problem of incentives.

A metric is useful because it stands in for something harder to observe, such as reliability, learning, or customer confidence. But any proxy leaves something out. When people optimize the visible part, neglected dimensions can deteriorate even as the dashboard improves.

The effect does not require dishonesty. Staff respond rationally to what the organization rewards. A teacher judged only by test scores may narrow lessons; a developer measured by tickets closed may avoid difficult maintenance that prevents future failures.

Targets can still help. A clear threshold can focus attention on a genuine problem, and reliable measures can reveal trends. Risk grows when one imperfect number determines high-stakes rewards, or when the people measured cannot influence the conditions behind it.

A team might track both the time to close a customer issue and whether the customer needs to contact support again. Neither measure tells the whole story, but together they make it harder to mistake a fast handoff for a lasting solution.

A better system combines indicators and judgment. Pair speed with resolution quality, short-term output with later rework, and individual results with team context. Use measures to ask questions and investigate patterns, not to pretend that one score contains the whole job.

Before adopting a target, ask what behavior it will encourage, what work it makes invisible, and how someone could improve the score without improving the outcome. Include workers in that review because they often see the loopholes and constraints first.

Pilot the measure and compare it with real work, user experience, and side effects. Revise it when it distorts behavior.

Managers also need to separate diagnosis from evaluation. A low number might signal poor effort, but it might instead reveal understaffing, outdated tools, or an unrealistic target. Treating every gap as a personal shortcoming hides the system that produced it.

A low number might signal poor effort, but it might instead reveal understaffing, outdated tools, or an unrealistic target.

If you could change one workplace metric, what behavior would you want it to reward, and what would you still need to observe directly?