Goodhart's Law: When Metrics Stop Measuring What Matters

Introduction

Goodhart's Law is one of the most practical mental models for anyone who manages goals, teams, products, habits, or performance. The core idea is simple: when a metric becomes the target, it often stops being a good measure. People start optimizing the number rather than the thing the number was supposed to represent.

You can see this everywhere. A customer support team is told to reduce average response time, so agents reply quickly with shallow answers. A school is judged by test scores, so teachers end up teaching to the test. A writer focuses on traffic alone, so headlines get sharper while substance gets thinner. The metric improves, but the real outcome often does not.

That is what makes Goodhart's Law so important. It explains why smart measurement systems still drift into nonsense. A number that starts as a useful signal can become a distorted target once incentives attach to it.

What Is Goodhart's Law?

Goodhart's Law is usually summarized like this: when a measure becomes a target, it ceases to be a good measure.

A metric begins as a proxy. It stands in for something harder to observe directly. Test scores stand in for learning. Revenue stands in for business health. Daily active users stand in for product value. Reply speed stands in for customer care.

The problem starts when people are rewarded, punished, ranked, or managed mainly through that proxy. Once that happens, behavior shifts. Instead of asking, "How do we improve the real thing?" people start asking, "How do we make the metric go up?"

That shift matters because proxies are never perfect. They capture part of reality, not the whole thing. The more pressure you put on the proxy, the more people learn to game it, narrow themselves around it, or sacrifice everything it does not measure.

Goodhart's Law does not say measurement is bad. It says measurement changes behavior. If you forget that, you end up managing numbers while the underlying system quietly deteriorates.

Why Goodhart's Law Matters

Many bad decisions start with good intentions. Leaders want accountability, so they choose metrics. Teams want clarity, so they define goals. Individuals want progress, so they track habits. None of that is wrong.

The trouble is that metrics are powerful incentives. Once you attach status, money, pressure, or identity to a number, people respond to the number. Sometimes that response helps. Often it creates side effects.

Goodhart's Law matters because it helps you notice three important truths:

• metrics shape behavior, not just observation
• a clean dashboard can hide a messy reality
• the most damaging distortions often look like success at first

This is why the model shows up in business, education, public policy, medicine, recruiting, finance, and personal productivity. Any system that relies on measurement is vulnerable.

It also connects closely to feedback loops. Once a metric becomes the target, the system starts reinforcing behaviors that improve the score. If the metric is weak, the loop magnifies the wrong thing.

How Goodhart's Law Works

The mechanics are straightforward.

1. You pick a metric because it roughly tracks something you care about.
2. You start rewarding people for improving that metric.
3. People adapt their behavior to improve the metric directly.
4. The metric becomes less connected to the real outcome.
5. You get better numbers and worse judgment.

The key issue is adaptation. People are not passive. Once measurement affects rewards, they search for the easiest path to the target.

Sometimes that means gaming. Sometimes it means cutting corners. Sometimes it means ignoring everything outside the metric. And sometimes it means optimizing locally while damaging the larger system.

That last point is especially important. A sales team can hit quarterly quotas by pushing bad-fit customers. A content team can raise pageviews with cheap clickbait. A recruiter can shorten time-to-hire by lowering quality standards. In each case, the metric improves while the wider system gets worse.

A Simple Example: Customer Support

Imagine a company decides to improve customer support by tracking one number: average first response time.

At first, this seems sensible. Faster responses usually feel better for customers. But once the team is judged mostly on that number, the behavior changes.

Agents start sending quick placeholder replies like "We're looking into this" just to stop the timer. Harder tickets get split into shallow updates. Complex issues get bounced around because each handoff protects the local metric. The dashboard looks great. Customers still feel frustrated.

What happened?

The company confused a proxy with the outcome. Fast first response can matter, but it is not the same thing as resolution quality, trust, or customer satisfaction. The metric was useful until it became the main target.

This is Goodhart's Law in action. The measure stopped measuring what mattered because the system learned to serve the measure.

Why Metrics Get Distorted

Metrics usually break in one of a few predictable ways.

1. Narrow optimization

People focus so tightly on one number that they ignore other important variables. If a writer chases word count, quality may fall. If a factory chases speed, defects may rise. If a teacher chases test scores, deeper understanding may shrink.

2. Gaming

People improve the metric without improving the underlying reality. This can be deliberate or unconscious. A team may redefine what counts, change the timing window, or move work off the books.

3. Proxy drift

A metric that once correlated with the real outcome becomes less useful over time because the environment changes. A growth metric may work when a product is new, then become misleading when the market matures.

4. Local optimization

One part of the system improves its own number by pushing costs onto another part. Procurement lowers costs by buying lower-quality inputs. Support lowers ticket counts by making it harder to contact support. Each team wins locally while the organization loses globally.

These are different failure modes, but they all point to the same lesson: no metric stays trustworthy under pressure unless the surrounding system is designed with care.

Real-World Examples of Goodhart's Law

Goodhart's Law becomes easier to remember when you attach it to concrete cases.

Education

If school quality is judged mainly through standardized test scores, classrooms may drift toward test preparation instead of real learning. Students get better at passing a narrow assessment, but not necessarily at understanding, writing, reasoning, or applying ideas outside the exam.

Hiring

If recruiting teams are measured mostly on speed, they can fill roles faster by reducing standards or by choosing low-risk candidates who look familiar. Time-to-hire improves while long-term team quality weakens.

Healthcare

If clinicians are judged by how many patients they see per hour, throughput may increase while listening, diagnosis quality, and trust decline. The metric captures volume but misses the human outcome.

Social Media

If creators optimize entirely for clicks or watch time, content often becomes more sensational, polarizing, or addictive. Engagement rises, but informational value, trust, and long-term brand quality may fall.

Personal Productivity

If you track only how many hours you worked, you may end up rewarding visible effort over meaningful progress. The number goes up. The important project still does not move.

Across all these examples, the same structure appears: the metric is not wrong, but it is incomplete. Once it becomes the target, people learn to bend themselves around its blind spots.

Goodhart's Law vs. Healthy Measurement

It is easy to overreact and conclude that all metrics are traps. That would be a mistake.

Metrics are useful because reality is noisy. You often need indicators to see progress, compare options, and identify problems early. The goal is not to stop measuring. The goal is to measure intelligently.

Healthy measurement usually has a few traits:

• it uses multiple signals instead of one dominant number
• it checks outcomes, not just activity
• it combines quantitative metrics with human judgment
• it gets reviewed when behavior starts drifting
• it stays tied to the real objective

This is similar to margin of safety. You should assume your metrics are imperfect and build slack into the system. If you rely too heavily on one proxy, small distortions can create large strategic mistakes.

Common Mistakes When Applying Goodhart's Law

The model is useful, but people misuse it in predictable ways.

Mistake 1: Treating one metric as the whole truth

A metric can be helpful without being complete. Problems begin when one number dominates decision making and crowds out everything else.

Mistake 2: Measuring what is easy instead of what matters

Organizations often choose clean metrics because they are available, not because they are meaningful. What is easiest to count is rarely the full story.

Mistake 3: Ignoring incentives

A dashboard is not neutral. If the people being measured have strong reasons to improve a score, they will adapt. Any measurement system that ignores incentives is fragile by design.

Mistake 4: Forgetting second-order effects

Improving the metric may create hidden costs elsewhere. This is where second-order thinking helps. Ask not only whether the number improves, but what behaviors the system rewards over time.

Mistake 5: Using lagging metrics too late

Sometimes by the time the metric clearly worsens, the damage is already done. Trust, culture, brand quality, and learning quality often degrade before the dashboard catches up.

How to Use Goodhart's Law in Practice

You do not need to become cynical about measurement. You just need a better process.

Start with the real objective

Before choosing a metric, ask: what outcome are we actually trying to produce? The clearer the objective, the easier it becomes to spot weak proxies.

Use a basket of metrics

One number is fragile. A small set of complementary measures is stronger. If you track customer support, for example, combine response time with resolution rate, satisfaction, repeat contacts, and qualitative review.

Watch behavior, not just numbers

If a metric improves suddenly, ask what changed in behavior. Better results may reflect a real improvement, but they may also reflect gaming, narrowing, or a shifting definition.

Review the metric regularly

A useful metric today may become misleading later. Revisit whether it still tracks the real outcome under current incentives and conditions.

Add qualitative judgment

Numbers compress reality. That is useful, but dangerous if taken too far. Good systems leave room for observation, sampling, and context.

Ask the inversion question

Borrowing from inversion, ask: if we wanted people to game this metric, how would they do it? That question often reveals blind spots immediately.

A Practical Checklist for Better Metrics

When you are choosing or reviewing a metric, these questions help:

1. What real-world outcome is this metric supposed to represent?
2. What behaviors will people adopt if they are judged by this number?
3. How could the metric be gamed or improved without real progress?
4. What important outcomes does this metric ignore?
5. Which second metric would balance its blind spots?
6. How often should we review whether it still measures what matters?

This checklist turns Goodhart's Law from a warning into a design tool. Instead of just distrusting metrics, you learn to build better ones.

Final Thoughts

Goodhart's Law is a reminder that measurement is never just measurement. The moment a metric becomes a target, it starts shaping behavior, incentives, and tradeoffs. If the metric is too narrow, the system bends around it and the number becomes less meaningful.

That is why good judgment requires more than dashboards. You need to stay connected to the real objective, watch for distortion, and keep asking whether the score still reflects reality. Used this way, Goodhart's Law helps you protect what matters instead of accidentally optimizing it away.

If you want a deeper framework for using mental models in everyday decisions, 100 Mental Models expands on these ideas in a broader and more practical way.