Designing Metrics

Designing Metrics That Don't Lie

The one idea: Any number a team is rewarded for will get pushed on — including the ways that make the number rise without the underlying goal improving. So never track a single metric alone. Pair it with the true outcome it's standing in for, and with guardrails that catch the shortcuts.

⚠️ The trap: "Closures jumped 25% after we set the target — the team is crushing it." That feels right and it's wrong. **The moment a measure becomes a target it stops measuring the real thing (Goodhart's law): here the headline rose while durably resolved problems actually fell, because nothing was watching how the number went up — pair every metric with a guardrail.**

The support team hit its number. Management had asked for 25% more tickets closed, and the team delivered: closures went from 800 a week to 1,000, bonuses were paid, the slide had a green arrow. Then the quarterly numbers came in and customer satisfaction had fallen — from 82% to 71% — and repeat contacts had climbed 3.75× (from 80 to 300 a week, nearly quadrupling). Nobody lied. Nobody cheated, exactly. The team did precisely what it was measured on, and doing it made customers worse off.

This is the central hazard of metric design. A metric is a proxy — a stand-in for something you actually care about but can't measure directly. "Tickets closed" stands in for "customer problems solved." As long as those two move together, the proxy is useful. The moment a team can move the proxy without moving the real thing, the metric stops describing reality and starts describing effort spent on…