The wrong metric will quietly corrupt a good strategy. People optimize what you measure, so a poorly chosen number pulls effort toward the number and away from the goal it was meant to represent. Read this and you will know how to pick metrics that stay honest, spot the ones already going bad, and protect your strategy from measuring itself into failure.
Why a metric drifts from its purpose
The underlying trap has a name. Goodhart’s law, associated with the economist Charles Goodhart and sharpened by the anthropologist Marilyn Strathern, holds that when a measure becomes a target, it ceases to be a good measure. The moment people are rewarded for a number, they optimize the number, including in ways that damage the outcome it stood for.
The cause is simple. Every metric is a proxy. “Customer satisfaction” is real but abstract, so you measure a survey score instead. The score is not satisfaction; it is a shadow of it. Push hard enough on the shadow and people learn to move the shadow without moving the object, by timing surveys, nudging responses, or excluding unhappy customers.
The difference between a signal and a target
A metric used as a signal informs judgment. A metric used as a target commands behavior. The same number behaves very differently depending on which role you give it. Signals can be softer and more numerous. Targets must be chosen with far more care, because they will be gamed.
Traits of a metric that resists corruption
- It is close to the real outcome, not a distant proxy.
- It is hard to move without also moving the thing you care about.
- It is paired with a counter-metric that catches the obvious gaming.
- It measures a result, not merely activity.
Pair every target with a counter-metric
The strongest defense is to never let one number stand alone. Pair it with a metric that gets worse if the first is gamed. Speed of resolution pairs with reopened-ticket rate. Sales volume pairs with refund or churn rate. When gaming one number degrades its partner, the incentive to cheat collapses.
| Primary target | How it gets gamed | Counter-metric |
| Support tickets closed per hour | Closing without resolving | Reopened-ticket rate |
| New signups | Low-quality or fake accounts | 30-day active rate |
| Lines of code shipped | Bloated, low-value code | Defect rate and rework |
A real scenario
A support team was measured on average handling time. The number improved beautifully. So did complaints. Agents were closing tickets fast by ending chats before problems were solved, then letting customers reopen new ones. The metric said the team was excellent. The customers said otherwise. Adding a reopened-ticket counter-metric fixed the incentive in a single quarter, because closing early now visibly hurt the second number.
Common mistakes and how to fix them
- Measuring activity instead of outcome. Calls made, hours logged, features shipped. Fix: tie the metric to a result someone outside the team would value.
- Using one number with no counterweight. Fix: pair every target with a metric that degrades if the first is gamed.
- Never retiring a metric. Metrics that were useful can go stale or start distorting. Fix: review each target and ask if it still reflects the goal.
- Tying compensation to a fragile proxy. Money amplifies gaming. Fix: reserve pay incentives for outcomes that are genuinely hard to fake.
Action steps
- For each key metric, write the real outcome it is standing in for.
- Ask: how would a clever, lazy person move this number without doing the real work?
- Add a counter-metric that catches that shortcut.
- Separate signals (for judgment) from targets (for incentives), and use fewer targets.
- Review every target quarterly; retire the ones that have drifted.
- Be cautious before attaching pay to any single proxy.
Conclusion and next step
A metric is a servant of strategy, not a substitute for it. The goal is measurement that stays honest under pressure. Your next step: take your single most important metric and ask how you would game it if you were paid to. The answer tells you exactly which counter-metric you are missing.
FAQ
Does Goodhart’s law mean I should not use metrics?
No. It means you should not blindly turn every metric into a hard target. Use numbers to inform judgment, choose targets carefully, and pair them with counter-metrics.
How many metrics should a strategy have?
Few targets, more signals. A handful of well-chosen targets is easier to keep honest than a long dashboard where each number quietly competes with the others.
What makes a good counter-metric?
One that gets worse precisely when the primary metric is gamed. If someone cheats the first number and the second stays flat, your counter-metric is not doing its job.
Should I tie bonuses to metrics?
Only to outcomes that are hard to fake and clearly tied to real value. Attaching money to a fragile proxy is the fastest way to trigger the exact gaming Goodhart’s law predicts.
References
- Charles Goodhart, origin of Goodhart’s law; Marilyn Strathern, its widely cited formulation.