Every quality professional eventually watches a measurement system fail. The instruments maintain their calibration. The operators hold their training certifications. The ISO 9001 and IATF 16949 documentation is pristine. Yet the metrics driving the management reviews no longer reflect the reality on the shop floor.
I have audited plants where the first-pass yield hit 99.8%, right before a major customer line-down event. The dashboards flashed green while the 8D reports piled up. The underlying processes had decayed, but the numbers looked spectacular because the organization had learned to optimize the metric rather than the process.
This decay follows a predictable, systemic pattern. Charles Goodhart formulated the original economic principle in 1975, which was later distilled by Marilyn Strathern: when a measure becomes a target, it ceases to be a good measure. Goodhart's Law actively erodes every KPI matrix if system designers fail to account for it.
The Anatomy of Metric Decay
Metric decay in a manufacturing environment happens in four distinct phases. Phase one introduces a genuinely useful indicator, like overall equipment effectiveness (OEE) or first-pass yield. For the first quarter, the indicator works. Decisions improve, waste drops, and the numbers rise because actual quality improves. The data accurately reflects process capability.
In phase two, leadership elevates the stakes. They tie the metric to performance reviews, bonus structures, or departmental rankings in IATF 16949 management reviews. The metric stops being a passive observation tool and becomes an active source of financial and professional consequences for the team. The relationship between the observer and the data changes.
Phase three is where the gaming begins. Human beings are resourceful when their livelihoods depend on a specific output. Boundary conditions in the control plan get reinterpreted. Borderline dimensions suddenly pass inspection. Rework gets reclassified as standard process variation.
By phase four, the metric is dead. The dashboard shows a flawless Cpk of 2.0, but warranty claims and internal scrap costs remain stubbornly flat. The metric has stopped measuring process stability and has started measuring the organization's ability to manipulate data entry.
The Four Phases of Metric Decay
- 011. The Metric WorksIndicator is introduced; decisions improve as data reflects actual process capability.
- 022. The Stakes RiseNumbers are tied to bonuses or rankings; information becomes a lever for consequences.
- 033. The Gaming BeginsRational actors reinterpret boundary conditions and reclassify defects to survive.
- 044. The Metric DiesKPI looks perfect while real quality flatlines; the number now only measures manipulation.
Patterns of Metric Corruption on the Shop Floor
Over two decades of auditing automotive and aerospace facilities, I have observed metric corruption manifest in distinct, repeated patterns. The most common is simple reclassification. A Tier 1 automotive supplier might report 99.4% on-time delivery by categorizing late shipments as "rescheduled by customer" after pressuring the client to accept the delay. The KPI measures the commercial leverage used on the customer, not the logistical capability of the plant.
Another frequent pattern is threshold fixation. If leadership sets a scrap rate target of 2%, the team will optimize to hit exactly that number. When scrap runs low early in the month, operators may ease off on tight process controls to avoid "wasting" their monthly buffer. This is not dishonesty; it is rational behaviour given the incentive structure.
Organizations also suffer from measure migration. A medical device manufacturer I advised obsessively tracked a final inspection pass rate of 99.1%. What went unmeasured was the explosion of non-conformances caught at incoming inspection. The organization had abandoned process control for end-of-line sorting. The output metric improved, but the cost of poor quality tripled.
The Cost of Aggregation and Context Stripping
Data smoothing routinely hides critical failures in aerospace facilities. A monthly customer complaint chart might show a beautiful, smooth decline. When you demand the raw weekly data, a different picture emerges: extreme volatility with systematic spikes every Tuesday following weekend changeovers.
Aggregating data to the level where it looks presentable on a slide destroys the resolution required to actually manage the process. The reporting cadence artificially smooths the variation, masking the exact instability the quality system is supposed to expose.
This is compounded by target inflation without context. A plant manager celebrates improving OEE from 72% to 86%. The reality? Maintenance successfully lobbied to reclassify three hours of preventative maintenance as "planned downtime," removing it from the denominator. The physical equipment ran exactly the same number of hours, but the calculation shifted.
The same applies to complaint reduction metrics. If customer complaints drop by 40% after you implement an automated phone menu that discourages callers, your product quality did not improve. You simply built a higher barrier to reporting. Goodhart's Law rewards this behaviour if the system architecture permits it.
System Design: Why This Keeps Happening
Goodhart's Law is not a failure of operator integrity. It is a failure of management system design. When you attach consequences to a metric, you change the dynamic between the observer and the observed. The act of measurement alters the behaviour being measured.
Most plants operate under what I call metric fundamentalism: the belief that the measurement is the thing itself. But a PFMEA score or a Cpk value is not the physical part. These metrics are shadows cast by the actual manufacturing process. Chasing the shadow does not improve the process.
Outcome Metrics vs Process Metrics
What teams track (Outcomes)
- Final scrap rate percentage
- Monthly customer complaint count
- Final inspection first-pass yield
- Overall equipment effectiveness output
What works (Processes)
- Control plan adherence rate
- 8D corrective action cycle time
- Statistical process control triggers
- Preventative maintenance completion
There is a critical cognitive shift when people are evaluated purely on numbers. Their mental bandwidth shifts from figuring out how to improve the process to figuring out how to improve the number. These sound like identical goals, but they require fundamentally different actions and yield drastically different operational results.
Designing a Goodhart-Resistant Quality System
You cannot eliminate metric decay, but you can design quality systems that resist it. The most effective defence is decoupling metrics from individual performance evaluation. Use dashboards as flashlights to illuminate problems, not as scoreboards to judge departments. The moment you tie a supervisor's bonus to first-pass yield, you plant the seed of corruption.
Accountability must instead rely on triangulation. One metric can be manipulated, but three independent data sources that converge on the same reality are incredibly difficult to fake. If you want to measure quality, examine defect rates, customer warranty claims, internal audit findings, and employee turnover in quality-critical roles simultaneously.
Rotate your investigative metrics while maintaining a stable core of health indicators. Treat your KPI suite like a medical checkup. You always monitor blood pressure and heart rate (core metrics like on-time delivery and scrap). But a doctor rotates in specific tests based on symptoms.
Finally, separate measurement from management. The person who measures the process should not be the person evaluated by it. In robust AS9100 or IATF 16949 systems, the data collection function is independent. The production team runs the line; the quality team monitors the data independently. Neither reports to the other.
What gets measured gets managed, and whatever you attach consequences to gets gamed.
The Culture That Defeats Bad Data
The most dangerous sentence in quality management is the persistent misattribution to Peter Drucker that "what gets measured gets managed." It is dangerously incomplete. The full operational truth is that whatever you attach consequences to will be gamed by the people facing those consequences.
The facilities with the best quality cultures do not possess the most sophisticated dashboards. They hold the most rigorous conversations about their metrics. When a yield drops or scrap rises, effective leaders do not demand to know who is responsible for the failure. They ask the team what the data is trying to tell them about the process.
Your metrics will always lie to you if you ask them to validate your excellence. Treat them as imperfect windows into a complex, variable reality. A Cpk value or OEE percentage is not the thing itself. The number merely points toward the thing. Building a system that respects that distinction is the only way to maintain genuine process control.