Goodhart’s Law and the Weaponization of KPIs

What is Goodhart’s Law and how does it apply to KPIs?

Goodhart’s Law predicts that any metric used as a target will be optimized at the expense of the underlying reality it was designed to measure, transforming KPIs from diagnostic instruments into performance theater.

The first time I saw this happen clearly was in a scheduling operation where “programs delivered” was the primary KPI. The target was 1,000 programs per year. The team hit the target by splitting 600 substantive programs into 1,000 smaller units, each technically countable as a delivered program. The dashboard showed green. The actual programming quality, measured by student engagement and completion rates, declined by 18%.

The metric had not failed. It had been optimized. The people optimizing it were not dishonest. They were rational actors responding to the incentive structure that the metric created. This is the essential insight of Goodhart’s Law: the problem is not bad actors gaming good metrics. The problem is that the act of targeting any metric inevitably creates incentives to optimize the metric at the expense of the thing it represents.

How do organizations weaponize KPIs?

Organizations weaponize KPIs by selectively choosing which metrics to report, setting targets that incentivize metric optimization over genuine improvement, and using dashboard visibility as a proxy for operational health.

I have built and maintained reporting dashboards for operations spanning 20,000+ annual student interactions. The patterns of metric weaponization are consistent:

• Cherry-picking: Report the 3 metrics that show improvement. Suppress the 7 that show decline. I built a dashboard that displayed 14 operational metrics. Leadership requested it be reduced to 4. The 4 selected were the most favorable, not the most diagnostic.
• Denominator manipulation: A “95% completion rate” can mean 95 of 100 students finished the program, or it can mean 95 of 100 students who were still enrolled at the midpoint finished. The difference is the 30 students who dropped before the midpoint and were quietly removed from the denominator.
• Temporal gaming: Shift activities across reporting boundaries to smooth numbers. I observed a pattern where program registrations were delayed to inflate the following quarter’s growth rate. The annual total was unchanged. The quarterly narrative was fabricated.
• Proxy substitution: When the real outcome is hard to measure (student learning, program impact, operational resilience), organizations substitute the nearest measurable proxy (attendance, survey scores, volume) and gradually forget that the proxy is not the thing itself.

What makes a KPI resistant to Goodhart corruption?

A KPI resists Goodhart corruption when it measures outcomes that are genuinely difficult to manipulate without actually improving the underlying system, and when it is paired with counter-metrics that detect the most common gaming strategies.

After watching several rounds of metric optimization that degraded actual performance, I adopted a set of principles for metric design:

• Pair every target metric with a counter-metric: If the target is “programs delivered,” the counter-metric is “student completion rate.” If the target is “records processed,” the counter-metric is “validation error rate.” The counter-metric should move in the wrong direction if the target metric is being gamed.
• Measure latency, not just volume: Volume metrics (records processed, programs delivered, tickets resolved) are the most gameable because they respond to unit-splitting. Latency metrics (time to resolution, processing cycle duration) resist this manipulation because they measure the speed of the actual work, not its fragmentation into countable units.
• Rotate metrics periodically: A metric that has been the primary target for 12+ months has almost certainly been partially corrupted. Rotating the primary target refreshes the measurement while the previous target becomes a diagnostic indicator stripped of its incentive distortion.

What does the weaponization of KPIs reveal about organizational epistemology?

The weaponization of KPIs reveals that most organizations operate not on the basis of truth but on the basis of narrative, using metrics as rhetorical devices to support predetermined conclusions rather than as instruments for discovering uncomfortable realities.

The philosopher Karl Popper argued that genuine knowledge requires the possibility of falsification. A theory that cannot be proven wrong is not scientific; it is ideology. By the same logic, a KPI system that cannot surface bad news is not a measurement system. It is a propaganda apparatus.

I have sat in meetings where every dashboard showed green while the people operating the systems described a workplace in crisis. The metrics were not lying. They were answering the questions they had been designed to answer, questions carefully selected to produce favorable responses. The data was accurate. The picture was false. This is the final, deepest consequence of Goodhart’s Law: it does not merely corrupt individual metrics. It corrupts the entire relationship between an organization and the truth about its own performance, replacing genuine understanding with the comfortable fiction that whatever the dashboard shows is whatever reality is.