CompanyGraph
A measure can describe performance while it is observed. Once rewards, penalties, or survival depend on the number, people change the activity—and sometimes the number becomes less informative.
What Goodhart's Law actually says
Goodhart's Law is not a mathematical rule that every target must fail. It is a warning about a change in use. A measure is selected because it correlates with a broader objective. When the measure becomes a target with consequences, the people being measured can alter what is counted, when it is counted, or which cases enter the measure. The original correlation may weaken.
The idea sits alongside Campbell's Law, which emphasizes that high-stakes indicators invite corruption or distortion of the process they monitor. The two are related but not identical. Goodhart's point concerns the loss of a statistical relationship when a measure is optimized; Campbell's concerns the pressure created by social decisions attached to the indicator. Neither implies that every target is useless or that every manager acts dishonestly.
The practical question is: what part of the desired outcome does the metric omit, and who can change the boundary around the number?
How a proxy becomes a target
Suppose a call centre wants customers to receive useful help. Average handling time is easy to record, so management sets a target. Agents now have a reason to end calls quickly, transfer difficult cases, or avoid work that increases the average. The number may improve while unresolved problems increase. If management later adds a repeat-call rate, the team can optimize both by changing coding or discouraging surveys. More measures reduce some blind spots but create a larger system with new boundaries to manage.
There are several routes to divergence:
- Selection: people choose which customers, patients, or cases are included.
- Timing: work or costs are moved across reporting periods.
- Substitution: effort shifts toward the measured task and away from an unmeasured one.
- Definition: categories or thresholds are interpreted to improve the count.
- Externalization: a burden is moved to a supplier, later period, or different team.
A documented banking case
The 2016 Wells Fargo account scandal provides a concrete boundary. The Consumer Financial Protection Bureau's enforcement action described employees opening unauthorized deposit and credit accounts to meet sales targets. The account count was observable, but it no longer represented willing customer relationships. The case involved prohibited conduct, yet the analytical lesson is broader: a target tied to employment and compensation changed the process that generated the measure.
The case does not prove that all cross-selling metrics are invalid. It shows why an investor should ask whether account growth is accompanied by funded balances, active use, retention, customer complaints, reversals, and economics after servicing costs. A target can be met while the underlying relationship is absent.
Where the law does not apply cleanly
Some targets improve the underlying outcome because the measured action is close to the objective. A factory's defect count may fall when a well-designed inspection process catches causes early. A subscription renewal rate may reflect real customer value when cancellation is easy and usage is verified. The metric becomes less reliable when the target changes the population, the process, or the cost borne elsewhere.
Metrics can also move for reasons unrelated to gaming: a new product mix, a change in customer composition, a recession, or a genuine process improvement. An investor should not label every divergence between indicators as manipulation. The hypothesis needs evidence about incentives, decision rights, timing, and the unmeasured result.
One divergence between reported profit and cash has a live screen: companies reporting positive margins while depreciation runs large against operating cash flow and receivables have grown four years in a row.
Earnings Growth With Heavy Accrual Component
Net profit margin is positive while depreciation is large relative to operating cash flow and receivables have grown four years in a row
An accrual-heavy profile is a prompt to follow the numbers across statements, not evidence of manipulation. Mix, investment timing, and genuine growth can produce the same shape.
How to audit a target system
- Define the objective in observable terms. “Customer quality” or “efficiency” is too broad. Name the outcome, time horizon, and people affected.
- Map who controls the numerator and denominator. Can the team choose cases, defer costs, change definitions, or exclude failures?
- Pair the target with a cost or quality measure. Revenue needs cash collection and retention; speed needs rework and repeat demand; margin needs service quality and asset condition.
- Look for discontinuities. A sudden improvement after compensation changes, a threshold effect just below a target, or a divergence from customer outcomes deserves investigation.
- Check the response loop. Who reviews exceptions, hears complaints, and can change the target? A dashboard without corrective authority only documents the drift.
What investors can infer
Goodhart's Law is most useful as a question about measurement design. It does not say that a company is fraudulent because it reports a target. It says that the target changes incentives and therefore changes the conditions under which the target should be interpreted.
When related measures move together across different incentives and time periods, confidence in the underlying improvement rises. When one number improves while cash, retention, safety, quality, or customer behaviour deteriorates, the number has become a weaker proxy. The investor's task is not to find a perfect metric. It is to keep the metric connected to the outcome it was meant to represent.