The Warning From an
Economist
In 1975, the British economist Charles Goodhart wrote a paper about
monetary policy in the United Kingdom. Buried inside it was an
observation so precise and so devastating that it would eventually bear
his name. Goodhart’s Law, as paraphrased by Marilyn Strathern in 1997,
states: “When a measure becomes a target, it ceases to be a good
measure.”
Goodhart was writing about central banking. He noticed that whenever
the Bank of England tried to use a particular economic indicator to
guide policy, people would adapt their behavior to game that indicator,
and it would stop reflecting the underlying reality it was supposed to
capture. The number still looked like the number. It just didn’t mean
anything anymore.
If you work in quality or manufacturing, you have seen this law in
action more times than you can count. The defect rate that was supposed
to reflect product quality became a number that people learned to
manipulate. The on-time delivery metric that was supposed to measure
customer satisfaction became a deadline that justified cutting corners.
The first-pass yield that was supposed to indicate process capability
became a target that incentivized reclassifying defects out of
existence.
The metric did not change. The behavior around it did. And the
quality you were trying to measure quietly walked out the door while
everyone was busy celebrating the number.
The Anatomy of Metric
Corruption
Goodhart’s Law does not strike overnight. It is a slow, almost
imperceptible corruption that follows a predictable pattern.
Understanding that pattern is the first step to breaking it.
Phase One: The Honest Metric. Someone — usually a
quality manager or a process engineer — identifies a measurement that
genuinely reflects something important. Defects per unit. Cycle time.
Customer complaints per thousand shipped. At this point, the metric is
descriptive. Nobody is being rewarded or punished based on it. It is
simply information, and it is reasonably accurate because there is no
incentive to distort it.
Phase Two: The Target. Leadership discovers the
metric and decides to “manage by it.” Targets are set. Bonuses are tied
to performance. Dashboards are installed on the factory floor. The
metric is no longer just information — it is now a judgment, and
judgments have consequences. This is the moment Goodhart’s Law
activates. The switch from observation to incentive changes everything,
even though the number itself has not changed at all.
Phase Three: The Optimization. Human beings are
optimization engines. When you attach consequences to a number, people
will find the most efficient path to making that number look good.
Sometimes that path is genuine improvement. Often, it is not. Inspectors
who are measured on defects found will find more defects — including
borderline ones that never mattered before. Operators who are measured
on throughput will move faster — and skip steps that nobody is
measuring. Suppliers who are measured on on-time delivery will ship on
time — and quietly compromise on specifications that are not being
tracked this quarter.
Phase Four: The Decoupling. This is where the real
damage happens. The metric and the reality it was supposed to represent
begin to diverge. The defect rate drops to 0.3 percent, but warranty
claims stay flat. The first-pass yield climbs to 98 percent, but rework
hours do not decrease. Customer satisfaction scores are excellent, but
repeat orders are declining. The numbers tell a story of quality. The
factory tells a different story. Nobody knows which one to believe, so
they believe the numbers, because the numbers are on the dashboard and
the dashboard does not lie.
Except that, by Phase Four, the dashboard is the only thing that does
lie.
Manufacturing
Examples You Have Lived Through
You do not need theoretical examples. You have lived these.
The Zero-Defect Target. A plant manager announces
that the facility will achieve zero defects by the end of the quarter.
Inspectors, whose performance reviews now include defect rates, begin
classifying borderline products as conforming. Rework stations are
quietly relabeled as “reprocessing” and removed from the defect tracking
system. Products that would have been flagged in January are shipped in
March. The defect rate hits zero. The customer return rate doubles. But
the dashboard is green, and the plant manager gets a bonus.
The OEE Trap. Overall Equipment Effectiveness is a
powerful metric when used honestly. But when OEE becomes a target tied
to promotions and bonuses, magical things happen. Downtime that used to
be scheduled for preventive maintenance is deferred because it hurts the
number. Changeover times are reported as production time. Minor stops
under five minutes are reclassified as “micro-pauses” and excluded from
the calculation. OEE climbs from 72 percent to 85 percent. Actual
throughput does not change. Equipment failures increase. But the number
went up, and the number is what matters.
The Supplier Scorecard Spiral. A company implements
a supplier scorecard with weighted metrics: quality, delivery, cost,
responsiveness. Suppliers quickly learn which metric carries the most
weight and optimize for that one. If delivery is 40 percent of the
score, they will ship on time even if they have to airfreight at a loss,
because losing points on delivery costs more than the premium shipping.
Quality, weighted at only 20 percent, quietly erodes. The scorecard
numbers look better every quarter. The actual incoming quality gets
worse. The procurement team celebrates. The production team
despairs.
The Training Hours Mandate. An organization decides
that continuous improvement requires continuous training. They set a
target: every employee must complete 40 hours of training per year. By
year two, the training department is offering online modules that take
15 minutes but count for two hours. Employees log in, click through, and
get credit. The training hours metric exceeds the target. Actual
competence does not change. But the report looks excellent.
Why This Keeps Happening
Goodhart’s Law is not a failure of intelligence. It is a failure of
understanding human behavior.
The fundamental error is treating a metric as if it is the thing it
measures. A thermometer measures temperature, but it is not temperature.
If you start rewarding thermometers for reading 72 degrees, some of them
will start reading 72 degrees regardless of the actual temperature. The
thermometer has not changed. Your relationship to it has.
In manufacturing, we make this mistake constantly. We confuse the
defect rate with quality. We confuse the cycle time with efficiency. We
confuse the audit score with compliance. These are related but not
identical, and the gap between them is exactly where Goodhart’s Law does
its work.
There are three structural reasons this keeps happening in quality
organizations.
First, metrics are easier to manage than reality.
You cannot directly manage quality. You can manage the systems,
behaviors, and conditions that produce quality. But that is hard, slow,
and uncertain. Managing a number is fast, clear, and satisfying.
Leadership reaches for the lever that moves the dashboard because the
dashboard responds immediately. The factory floor does not.
Second, targets create information asymmetry. The
people closest to the work know things that the dashboard cannot
capture. They know which defects are real and which are classification
artifacts. They know which processes are genuinely improving and which
are just being measured differently. When their incentives are tied to
the metric rather than the reality, they have every reason to use that
information asymmetry to make the number look good. They are not being
dishonest. They are being rational actors in a system that rewards the
wrong thing.
Third, the time lag hides the damage. When you
corrupt a metric, the consequences do not show up immediately. The
defect rate improves this quarter. The customer complaints arrive next
quarter. The lost customers show up the quarter after that. By the time
the real damage is visible, the dashboard has moved on to new targets,
and nobody connects the current problems to the previous optimization.
The organization learns the wrong lesson: the metric improvement worked,
because look, the number went up.
The Countermeasures
Breaking Goodhart’s Law is not about abandoning metrics. It is about
changing your relationship to them. Here are the countermeasures that
actually work.
Measure Multiple Things
Never let a single metric dominate decision-making. If you are
tracking defect rate, also track warranty claims, customer returns,
rework hours, and scrap cost. If these metrics move together, the defect
rate is probably honest. If the defect rate improves but the others stay
flat or worsen, you are watching Goodhart’s Law in action and calling it
progress.
The power of multiple metrics is not precision. It is contradiction.
When metrics contradict each other, they reveal the truth that any
single metric would hide. Embrace the contradiction. Investigate it. It
is telling you something important.
Separate Measurement
From Consequences
This is the hardest countermeasure and the most important. If the
people who produce the metric are the same people whose careers depend
on it, the metric will be corrupted. Not because people are bad, but
because the incentive structure makes corruption rational.
The solution is structural independence. Quality auditors should not
report to the plant manager whose performance they are auditing.
Incoming inspection should not be funded by the procurement department
whose suppliers they are rejecting. Data collection should be automated
wherever possible, removing the human element from the measurement
itself.
This does not eliminate Goodhart’s Law entirely, but it raises the
cost of gaming the metric from “reclassifying a few borderline items” to
“actively falsifying records,” which most people will not do.
Rotate the Metrics
If you keep measuring the same thing the same way for years, people
will find the optimization path. The solution is to periodically change
what you measure and how you measure it. Not randomly — that would be
chaos — but deliberately, rotating through a portfolio of metrics that
all reflect quality from different angles.
One quarter, focus on first-pass yield. The next, focus on customer
returns. The next, focus on process capability indices. Each metric,
during its turn as the primary target, gets honest attention because the
optimization strategies from the previous quarter do not apply to the
new one. Over time, genuine improvement shows up across all of them,
because real quality improvement is not metric-specific.
Track the Delta
Between Metrics and Reality
Create a process for periodically auditing whether your metrics still
mean what they used to mean. Compare your internal defect rate to your
external customer complaints. Compare your OEE calculation to your
actual throughput. Compare your supplier scorecards to your incoming
rejection rate. When the gap widens, you know the metric is being
gamed.
This is essentially a meta-metric: a measurement of whether your
measurements are still valid. It sounds recursive, but it is the only
defense against the slow drift that Goodhart’s Law produces.
Ask the Operators
The people on the production floor know whether quality is actually
improving. They know whether processes are genuinely better or whether
the numbers are just being massaged. They usually will not volunteer
this information because there is no incentive to do so and considerable
risk in doing so. But if you create safe channels — anonymous surveys,
skip-level meetings, independent interviews — they will tell you the
truth.
Operators are your most underutilized quality metric. They are the
ones who see the gap between the dashboard and the factory floor every
single day. They are the canaries in the coal mine of Goodhart’s Law.
Listen to them.
The Deeper Lesson
Goodhart’s Law is ultimately about the limits of reductionism.
Quality is a complex, multidimensional reality. It lives in the material
properties of your product, in the skill of your operators, in the
reliability of your equipment, in the clarity of your specifications, in
the trust of your customers. It cannot be fully captured by any single
number, and any attempt to do so will eventually distort the very thing
you are trying to measure.
This does not mean metrics are useless. It means metrics are tools,
not truths. A caliper is a tool for measuring distance. It is useful
precisely because you understand its limitations — its resolution, its
accuracy, the conditions under which it reads true. Treat your quality
metrics the same way. Understand what they capture and what they miss.
Understand how they change when people know they are being watched.
Understand that the moment you attach a bonus to a number, you have
changed the number’s meaning even if you have not changed the number
itself.
The organizations that manage quality best are not the ones with the
most sophisticated dashboards. They are the ones that maintain a healthy
skepticism about their own measurements. They use metrics to ask
questions, not to provide answers. They treat a good number as a reason
to investigate further, not a reason to stop looking. They understand
that the map is not the territory, the menu is not the meal, and the
defect rate is not the quality.
Charles Goodhart was an economist writing about monetary policy in
1975. He could just as easily have been writing about your factory floor
today. The law he described is not specific to central banking. It is
specific to human beings. And until we stop being human, the law will
hold.
The metric you are not watching is the one that tells the truth. The
metric you are rewarding is the one that learns to lie. Plan
accordingly.
Peter Stasko is a Quality Architect with over 25
years of experience in manufacturing excellence, process optimization,
and quality systems design. He writes about the intersection of human
psychology and operational performance because he believes that the most
important variable in any quality system is the person running it.
