Quality and Goodhart’s Law: When Your KPIs Stop Measuring Quality and Start Measuring Your Ability to Game the KPIs
The Day a Perfect Dashboard Killed a Perfect Product
It was a Tuesday morning in a Tier 1 automotive plant in eastern Slovakia, and the monthly quality review was in full swing. The projector hummed. The slides glowed. Every chart was green. Every target was met. Scrap rate: 0.12%. Customer complaints: zero. PPM: 3. OEE: 91.7%. The plant manager leaned back in his chair with the quiet satisfaction of a man whose numbers told a story of excellence.
Then his phone rang.
It was the VP of Quality from their biggest customer — a German OEM whose contracts represented 40% of the plant’s revenue. The call lasted eleven minutes. When the plant manager hung up, his face had changed color.
Three of their parts had been found in a final assembly line in Munich with a dimensional deviation that should have been caught at incoming inspection. But the deviation was subtle — 0.03mm outside of spec on a critical mating surface. Subtle enough to pass a quick visual check. Subtle enough to assemble. Subtle enough to make it all the way to the end customer before the issue manifested as a noise complaint during a test drive.
The plant manager stared at his green dashboard. Everything was perfect. Everything was a lie.
What had happened wasn’t a failure of measurement. It was a failure of meaning. The KPIs weren’t measuring quality anymore. They were measuring the organization’s ability to make the KPIs look good.
This is Goodhart’s Law in action. And if you work in quality, manufacturing, or any domain where people are held accountable to numbers, it is silently shaping your reality right now.
What Is Goodhart’s Law?
The original formulation, attributed to British economist Charles Goodhart, is deceptively simple:
“When a measure becomes a target, it ceases to be a good measure.”
Marilyn Strathern later refined it with an anthropologist’s eye for human behavior:
“When a measure becomes a target, it ceases to be a good measure.”
The insight is this: the moment you attach consequences — bonuses, promotions, reputation, survival — to a specific metric, human beings will optimize for that metric. Not for the underlying reality the metric was supposed to represent. For the metric itself.
In economics, this means banks game lending ratios. In education, it means teachers teach to the test. In healthcare, it means hospitals admit the easy cases and avoid the sick ones.
In quality and manufacturing? It means your scrap rate drops to zero because your inspectors start reworking parts before they count them as scrap. Your customer complaint count stays flat because your sales team starts resolving complaints informally before they get logged. Your PPM looks beautiful because you recalculated the denominator.
None of this is fraud in the traditional sense. It’s something more insidious: it’s rational behavior in a system that rewards the wrong thing.
The Anatomy of Metric Corruption in Quality Systems
Goodhart’s Law doesn’t announce itself. It doesn’t show up as a sudden, dramatic failure. It creeps in through small, reasonable-sounding decisions made by well-intentioned people who are responding to the incentives you created.
Here’s how it typically unfolds in a manufacturing quality environment:
Phase 1: The Metric Is Honest
A new KPI is introduced. Let’s say it’s “first-pass yield” on a critical machining line. For the first few months, the number is genuine. The process either produces good parts on the first try or it doesn’t. The metric reflects reality. Decisions based on the metric are sound.
Phase 2: The Metric Becomes a Target
Leadership notices the yield number. It gets put on a dashboard. It gets discussed in meetings. Bonuses get tied to it. A yield target of 98% becomes the difference between a good quarter and a bad one.
At this point, the metric hasn’t changed. But its relationship to reality has. Because now people care about the number — not because it tells them something useful, but because it has consequences.
Phase 3: The Metric Begins to Drift
Subtle shifts start happening. The operator notices that if they run a part through the gauge twice and average the readings, the number improves slightly. The quality engineer notices that reclassifying a borderline defect as “cosmetic” rather than “functional” keeps the yield above target. The shift supervisor notices that if they stop the line for five minutes and restart, the yield counter resets.
None of these people are dishonest. They are solving the problem you gave them: make the number look good.
Phase 4: The Metric Is Meaningless
Six months later, the yield is 99.2%. It never drops below 98.5%. Leadership is pleased. But the underlying process hasn’t improved. The defect rate to the customer hasn’t changed. The rework area is as busy as ever. The only thing that changed is the distance between the number and the truth.
This is Goodhart’s equilibrium: the metric becomes a mirror of the effort to game the metric, not a window into the process it was supposed to describe.
The Seven Classic Goodhart Patterns in Manufacturing
Over two decades in quality, I’ve seen the same patterns repeat across industries, continents, and company sizes. Here are the seven most common manifestations:
1. The Scrap Rate Illusion
You set a scrap target. Scrap drops. But where did the defective parts go? In many organizations, they didn’t disappear — they were reclassified. “Rework” isn’t scrap. “Deviation” isn’t scrap. “Customer concession” isn’t scrap. The parts are still defective. They still cost money. They still represent process failure. But the scrap rate looks beautiful.
The tell: Scrap rate improves, but total cost of poor quality stays flat or rises.
2. The Customer Complaint Black Hole
Customer complaints are a key metric. So the organization learns to absorb complaints before they become formal. The quality engineer calls their counterpart at the customer and says, “Let us handle this — no need to log it.” The sales rep visits the customer with replacement parts and a box of chocolates. The complaint never enters the system.
The tell: Formal complaints drop, but warranty costs and customer churn increase.
3. The Audit Score Theater
Your internal audit scores are consistently above 95%. External audits find the same nonconformities year after year. The gap exists because internal auditors — who are evaluated on the scores they produce, or who work alongside the people they audit — have learned to see what the target requires them to see.
The tell: Internal audit scores don’t correlate with external audit findings or actual quality performance.
4. The Training Checkbox
Training completion rates hit 100%. Every operator has signed every training record. But when you walk the floor and ask an operator to explain the control plan for their station, they stare at you blankly. The training happened — on paper. The competence did not.
The tell: Training completion is 100%, but repeat defects from the same root cause persist.
5. The corrective action closure rate is 100%
Every corrective action is closed on time. But 30% of the same failures recur within six months. The actions were closed, not effective. The metric measured closure, not resolution.
The tell: CAPA closure rate is high, but repeat failures are constant.
6. The OEE Mirage
OEE is the holy trinity of manufacturing metrics: availability × performance × quality. But each component can be gamed. Availability improves by reducing scheduled maintenance. Performance improves by running faster than the process was designed for, creating quality problems that show up later. Quality improves by widening spec limits or reclassifying defects.
The tell: OEE improves while customer returns increase.
7. The Supplier Scorecard Conspiracy
You score your suppliers. Your scores look excellent — every supplier above 90%. But your incoming defect rate hasn’t changed. The scorecard has become a diplomatic tool rather than an assessment tool. Nobody wants to give a supplier a bad score because it means having to find a new one.
The tell: Supplier scores are uniformly high, but incoming quality is not.
Why Goodhart’s Law Is Particularly Dangerous in Quality
Quality systems are uniquely vulnerable to metric corruption for several reasons:
First, quality metrics are often lagging indicators. By the time a defect rate shows up in your data, the conditions that created it may have changed. This creates a temptation to manage the number rather than the process, because the process is harder to see.
Second, quality metrics are interconnected. Improving one metric often degrades another. Reduce scrap by increasing rework, and your throughput drops. Reduce customer complaints by accepting more concessions, and your specification integrity erodes. The moment you optimize for a single number, you lose sight of the system.
Third, quality metrics carry enormous emotional weight. A red number on a quality dashboard isn’t just information — it’s a judgment. It triggers defensiveness, justification, and the full spectrum of human avoidance behaviors. When people feel threatened by a metric, they don’t improve the underlying reality. They improve the metric.
Fourth, quality systems are built on trust. When Goodhart’s Law takes hold, trust erodes slowly and invisibly. The quality engineer who used to report bad news honestly starts hedging. The operator who used to pull the andon cord starts fixing problems quietly to avoid affecting the numbers. The culture shifts from “catch defects” to “manage the numbers.”
The Anti-Goodhart Framework: How to Keep Your Metrics Honest
You can’t eliminate Goodhart’s Law. It’s a feature of human nature, not a bug in your KPI system. But you can design systems that resist its effects.
Principle 1: Measure Systems, Not People
When a metric is tied to individual performance, gaming is inevitable. When it’s tied to system performance, the incentive shifts from “make my number look good” to “understand what’s actually happening.”
Instead of measuring operator defect rates, measure process capability. Instead of measuring inspector detection rates, measure defect escape rates at the system level. The people closest to the work should be your best source of truth, not your most motivated data manipulators.
Principle 2: Use Complementary Metrics
Every metric has a blind spot. Every blind spot is where Goodhart’s Law lives. The solution is to pair metrics that cover each other’s weaknesses.
If you measure scrap rate, also measure total rework hours. If you measure customer complaints, also measure warranty costs. If you measure OEE, also measure customer returns. When a single metric improves in isolation while its complement deteriorates, you’ve found Goodhart’s fingerprint.
Principle 3: Rotate What You Measure
Metrics that are measured consistently for years become games that are played consistently for years. Rotate your measurement focus periodically. This doesn’t mean abandoning important metrics — it means shifting which ones carry the weight of targets and consequences.
One effective approach is to have a core set of monitored metrics (no targets, just tracking) and a rotating set of targeted metrics (specific goals, specific timelines). When a metric rotates out of the target set and into the monitoring set, its true behavior is revealed — because the pressure to game it has been removed.
Principle 4: Separate Measurement From Consequence
The person who measures should not be the person who is measured. This is why third-party audits exist. This is why independent quality labs exist. This is why the most effective quality organizations have a measurement function that is structurally separated from the production function.
When the same person who is responsible for hitting a target is also responsible for reporting whether the target was hit, the reporting will drift toward the target.
Principle 5: Regularly Reconnect With the Underlying Reality
The most powerful antidote to Goodhart’s Law is direct observation. Go to the gemba. Watch the process. Talk to the operators. Compare what you see with what the dashboard says.
When a metric says the process is improving but your eyes tell you it isn’t, trust your eyes. The metric is measuring its own optimization, not the process.
Principle 6: Create Psychological Safety Around Bad Numbers
If people are punished for bad metrics, they will ensure the metrics are never bad — regardless of reality. The alternative is to create an environment where bad numbers are treated as information, not indictment.
This doesn’t mean abandoning accountability. It means redirecting accountability from the number to the response. You don’t hold people accountable for having a high defect rate. You hold them accountable for what they do about it.
The Deeper Lesson: Metrics Are Maps, Not Territory
Goodhart’s Law is ultimately a reminder of something we all know but conveniently forget: the map is not the territory. A KPI is a simplified representation of a complex reality. It captures some dimensions and ignores others. It compresses a dynamic system into a static number.
This simplification is necessary. You can’t manage what you can’t measure, and you can’t measure everything. But the simplification becomes dangerous the moment you forget that it’s a simplification — the moment you start treating the number as the reality rather than a shadow of the reality.
The best quality professionals I’ve worked with share a particular kind of skepticism. They look at their own dashboards with suspicion. They ask, “What is this number not telling me?” They know that every green light on a control panel could be a burned-out bulb.
The Slovakian plant manager from our opening story learned this lesson the hard way. After the customer crisis, he didn’t add more metrics. He did the opposite. He took the targets off the dashboard for three months and asked his team to simply observe and report what was actually happening — without consequences, without scores, without judgment.
What they found was a process that was producing 4.7% defective parts — most of which were being caught and reworked before they became scrap. The scrap rate of 0.12% was real. But it was measuring the rework system, not the production system.
The real yield was 95.3%. And until they saw that number, they couldn’t improve it.
A Final Word
Goodhart’s Law is not a flaw in your quality system. It is a feature of human cognition. People optimize for what they’re measured on. This is not cynicism — it’s realism.
The organizations that manage quality most effectively are not the ones with the most sophisticated KPI systems. They are the ones that maintain a healthy distrust of their own measurements. They use metrics as a starting point for inquiry, not as an ending point for judgment.
The next time you look at your quality dashboard and every number is green, ask yourself one question:
Is my process actually getting better, or have I just gotten better at making the numbers look the way they’re supposed to look?
The answer to that question is the most important metric you’ll never find on a dashboard.
Peter Stasko is a Quality Architect with over 25 years of experience in automotive and manufacturing quality management. He has led quality system implementations across multiple continents, guided organizations through IATF 16949 certification, and witnessed Goodhart’s Law in action more times than he can count — including in his own dashboards. He writes about quality not as a theoretical exercise, but as a craft practiced by real people on real shop floors, where the gap between the number and the truth is where all the interesting things happen.
