The
Law That Explains Why Your Quality Dashboard Lies to You
In 1975, the British economist Charles Goodhart articulated a
principle that would explain more quality management failures than any
defective process ever could: “When a measure becomes a target,
it ceases to be a good measure.”
Marilyn Strathern refined it later, and the result is what we now
call Goodhart’s Law. It is the most important concept in quality
management that almost no quality manager has ever heard of. And if they
have heard of it, they usually don’t believe it applies to them.
It does. It applies to you. Right now.
If your organization tracks first-pass yield, that number will go up.
If your organization tracks defect rates, that number will go down. If
your organization tracks on-time delivery, that metric will improve. And
the uncomfortable question — the one almost nobody asks — is whether
those improvements reflect actual quality gains or whether they reflect
something else entirely. Something darker. Something that happens when
human beings realize their performance reviews, their bonuses, and their
career trajectories are tied to a number they have the ability to
influence.
Goodhart’s Law describes the precise mechanism by which your quality
metrics become fiction. Not through fraud, not through conspiracy, but
through the slow, quiet, rational adaptation of intelligent people
responding rationally to the incentive structures you built around
them.
The Four Ways Your
Metrics Get Corrupted
Goodhart’s Law doesn’t operate through a single mechanism. It has
four distinct pathways, each of which destroys metric integrity in a
different way. Understanding these pathways is the first step toward
recognizing which one is active in your organization right now.
1.
Regressive Goodhart: Hitting the Target by Lowering the Bar
This is the most common and most insidious form. When you set a
target, people find the easiest way to hit it — and the easiest way
usually involves redefining what counts.
Your defect rate target is 2%. Your team starts classifying
borderline defects as “cosmetic variations” or “non-conformance
observations” rather than defects. The defect rate drops to 1.7%.
Leadership celebrates. The quality of the product hasn’t changed at all.
What changed was the classification system, quietly re-engineered to
make the number look better.
Your first-pass yield target is 95%. Your team starts routing
borderline units through a “reinspection and reclassification” loop that
technically removes them from the first-pass calculation. FPY hits 96%.
Nobody asks why the reinspection queue grew by 300%.
This isn’t cheating, exactly. It’s rational behavior in a system that
rewards the metric rather than the underlying quality. When you measure
something and attach consequences to the measurement, the measurement
itself becomes the work.
2.
Extremal Goodhart: Hitting the Target by Breaking Everything Else
When you optimize for one metric, you almost always deoptimize for
another — usually one you’re not measuring. This is extremal Goodhart,
and it’s particularly dangerous because the damage happens in your blind
spots.
Your organization decides to drive down cycle time. The team
complies. Cycle time drops 20%. What nobody measured was that the team
achieved this by cutting inspection steps, reducing setup thoroughness,
and pressuring operators to skip non-critical but quality-relevant
activities. Six months later, warranty claims spike. Customer complaints
triple. But the cycle time metric — the one on the dashboard, the one
tied to bonuses — still looks fantastic.
You optimized the metric. You destroyed the system. The metric became
a target, and the target became a weapon aimed at every unmeasured
dimension of quality.
3.
Causal Goodhart: Hitting the Target Through Means That Don’t Last
Sometimes a metric correlates with quality in normal conditions, but
the moment you push on it directly, the correlation breaks. This happens
because the metric was a symptom of good quality, not a cause of it —
and symptoms can’t be engineered backward into causes.
Employee engagement scores correlate with better quality output. So
leadership sets a target for engagement scores. Managers start holding
pizza parties, sending birthday cards, and creating a culture of
mandatory fun. The engagement scores go up — because employees learn
that low scores mean more mandatory fun, which is worse than no fun at
all. But actual engagement — the genuine commitment, intrinsic
motivation, and discretionary effort that drove quality in the first
place — doesn’t improve. It degrades. You measured a correlate, treated
it as a lever, and broke the correlation.
4.
Adversarial Goodhart: Gaming the System Deliberately
This is the form where Goodhart’s Law shades into active
manipulation. When stakes are high enough and metrics are gameable
enough, someone will eventually game them — not from malice, but from
the entirely rational calculation that gaming the metric is easier,
safer, and more rewarding than actually improving the underlying
process.
Your supplier quality rating is based on PPM defect rates. Your
supplier discovers that if they inspect and scrap defective units before
shipping, those units never appear in the PPM calculation. Their PPM
looks world-class. Your incoming inspection finds the same defect rate
it always did — but now you’re also paying for the supplier’s internal
scrap through higher unit prices. The metric improved. The cost went up.
The actual quality didn’t change.
The
Real-Body Count: Where Goodhart’s Law Lives in Your Factory
Let’s make this concrete. Walk onto your shop floor with Goodhart’s
Law as your lens and you’ll start seeing it everywhere.
OEE targets. You set an OEE target of 85%. Within
three months, OEE hits 86%. But look closer. Availability is calculated
from scheduled production time — and someone quietly redefined
“scheduled” to exclude changeovers and planned maintenance windows.
Performance rate uses ideal cycle time — and someone recalculated the
ideal cycle time to be 15% more generous. Quality rate excludes “rework”
from the defect count because rework isn’t technically a scrap event.
Every component of OEE has been individually re-engineered to produce a
better number. World-class OEE on paper. Same equipment, same process,
same actual output.
On-time delivery. Your OTD target is 98%. Your
customer service team starts calling customers three days before the
promised delivery date to “confirm” they’ll be available to receive the
shipment. If the customer can’t confirm, the delivery date gets
“rescheduled” — and now it counts as on-time because it was delivered on
the revised date, not the original date. OTD hits 98.5%. Your customers’
actual experience hasn’t changed.
Audit findings. Your internal audit program targets
fewer than five major findings per year. Auditors learn — through subtle
organizational pressure, not explicit instruction — that finding more
than five creates problems. Problems for the audited facility, problems
for the audit manager, problems for the auditor’s own performance
review. So audits become increasingly selective. Major findings become
minor findings. Minor findings become observations. Observations become
“opportunities for improvement” that don’t appear in the formal report.
The finding count drops to three. Leadership celebrates a quality
improvement that is, in reality, an audit quality degradation.
Training completion. You mandate 40 hours of quality
training per employee per year. Completion rates hit 97%. But the
training is a slide deck employees click through in 12 minutes while
doing their actual work on a second monitor. The metric says training
happened. The knowledge didn’t.
Why
Goodhart’s Law Is Not the Same as Laziness or Incompetence
Here’s the critical insight that most quality leaders miss:
the people corrupting your metrics are not your worst employees.
They are often your best.
Goodhart’s Law operates through intelligent, motivated people who are
responding rationally to the incentive system you designed. When you
attach consequences to a metric, you are implicitly telling your team:
“This number is what matters.” Your most capable employees — the ones
who are smart enough to understand the incentive structure and motivated
enough to act on it — will optimize for the metric you’ve chosen. That’s
not a character flaw. That’s exactly the behavior you engineered.
The problem isn’t that your team is lazy. The problem is that you
built a system that rewards metric optimization instead of quality
improvement, and then you were surprised when people optimized the
metric.
This is why adding more audits, more inspections, and more oversight
doesn’t fix the problem. You’re just adding more metrics to game, more
targets to corrupt, more layers of measurement that will themselves
become subject to Goodhart’s Law. You cannot measure your way out of a
measurement problem.
The
Diagnostic: How to Tell If Your Metrics Have Been Corrupted
You won’t find evidence of Goodhart’s Law by looking at the metrics
themselves — that’s the whole point. The corrupted metric looks great.
That’s how it got corrupted. Instead, look for these diagnostic
signals:
Divergence between metrics and outcomes. Your defect
rate is dropping, but your warranty claims are rising. Your FPY is
improving, but your customer complaints are increasing. Your OEE is
world-class, but your equipment failure rate is climbing. When your
metrics and your real-world outcomes tell different stories, Goodhart’s
Law is the narrator.
Metric improvement without process change. If your
metrics improved but you can’t point to a specific change in process,
technology, materials, or people that would explain the improvement, the
improvement isn’t real. Real quality improvement has a cause. Metric
corruption has only a motive.
Definitional drift. Compare how a metric was defined
when it was introduced to how it’s defined now. If the definitions have
been refined, clarified, adjusted, or “operationalized” in ways that
make the number look better, you’re watching Goodhart’s Law in slow
motion.
Concentration on the boundary. If your metrics
cluster suspiciously close to target values — always just above the
threshold, never dramatically better — someone is calibrating output to
the target rather than optimizing for genuine improvement. Natural
performance varies. Calibrated performance hovers.
Absence of bad news. If your quality dashboard has
shown consistent improvement for 18 months without a single anomaly,
reversal, or surprise, either you’ve achieved a level of process
perfection unknown in human history or your metrics have stopped telling
you the truth. Bad news is a sign that your measurement system still
works. The absence of bad news is the most alarming signal of all.
Designing
Around Goodhart’s Law: What Actually Works
You cannot eliminate Goodhart’s Law. It’s as fundamental as entropy.
But you can design measurement systems that resist it, detect it early,
and limit the damage.
Measure Diversely, Reward
Diversely
Never attach significant consequences to a single metric. Use
balanced scorecards that combine multiple, independent dimensions of
quality — dimensions that cannot all be gamed through the same
mechanism. If someone is corrupting one metric to hit a target, the
corruption will usually show up as an anomaly in a different metric.
The key insight: your metrics should be independent.
Two metrics that both improve when someone reclassifies defects are not
independent. Two metrics that both improve when someone redefines a
cycle time are not independent. Choose metrics that would require
contradictory forms of gaming to corrupt simultaneously.
Rotate Your Metrics
Don’t keep the same metrics in place for years. Rotate them. If the
organization knows that this quarter’s critical metric will be replaced
by a different one next quarter, the incentive to corrupt any single
metric drops significantly. You don’t have to change what you measure —
you have to change what you reward.
This is uncomfortable for organizations that want stable, comparable
metrics over time. But stable, comparable metrics over time are exactly
what Goodhart’s Law feeds on. The stability that makes your dashboard
look professional is the same stability that makes it gameable.
Separate Measurement from
Incentives
This is the nuclear option, and it’s the most effective. Have your
quality metrics measured by people who have no stake in the outcomes.
Independent measurement doesn’t eliminate Goodhart’s Law — the measurers
can still be influenced — but it inserts a layer of separation between
the people hitting the target and the people counting the hits.
Many organizations have moved quality reporting out of operations and
into a separate function for exactly this reason. The operations team
produces the quality. The independent quality function measures it. The
separation creates friction against corruption.
Track the Meta-Metrics
Monitor the measurement system itself. Track changes in definitions,
changes in classification criteria, changes in what gets excluded. Track
the gap between what your metrics say and what your customers say. Track
the ratio of reported improvement to actual investment in process
change. If your metrics are improving faster than your process
investment would predict, something besides quality is driving the
numbers.
Ask the Question Nobody Asks
The single most powerful defense against Goodhart’s Law is a question
that quality leaders almost never ask: “If our metrics are this
good, why isn’t our quality better?”
This question forces the conversation past the dashboard and into the
reality the dashboard is supposed to represent. If your defect rate has
dropped 40% but your customers haven’t noticed, either your customers
are unobservant or your defect rate measurement has drifted away from
the reality of defects. If your process capability indices have improved
dramatically but your scrap costs haven’t moved, your Cpk calculations
and your actual process capability have diverged.
The answer to this question is almost always uncomfortable. That’s
why nobody asks it. That’s exactly why you should.
The
Uncomfortable Truth About Your Quality Dashboard
Here’s what Goodhart’s Law is ultimately telling you: your
quality metrics are not a photograph of your quality. They are a
painting, and your organization is the artist.
The metrics you trust most — the ones that show consistent
improvement, the ones that always hit target, the ones that make your
quality program look successful — those are the metrics most likely to
have been corrupted. Not because your team is dishonest, but because
they’re human, and humans adapt to incentive structures. That’s not a
bug. That’s a feature of the species.
The quality leader who understands Goodhart’s Law approaches their
own dashboard with suspicion. Not because they don’t trust their team,
but because they understand that trust in people and trust in metrics
are different things entirely. You can have the most honest, dedicated,
capable team in the world, and your metrics can still be fiction —
because Goodhart’s Law doesn’t require dishonesty. It only requires
incentives, time, and intelligence.
Your metrics became targets. Your targets became games. Your games
became the quality you think you have.
The question is whether you have the courage to find out what your
quality actually is — and whether you’re willing to dismantle the
measurement system that makes you feel good in order to build one that
tells you the truth.
About the Author: Peter Stasko is a Quality
Architect with over 25 years of experience transforming quality
management systems across manufacturing organizations. He specializes in
separating metric theater from genuine quality improvement — because the
number that makes you comfortable is usually the number that’s lying to
you.
