When a measure becomes a target, it ceases to be a good measure.
This is Goodhart's Law. It applies devastatingly to retention metrics. Once you incentivize NRR, teams optimize NRR at the expense of what NRR was supposed to indicate.
How Goodhart's Law works
A metric starts as a proxy for something you care about. NRR is a proxy for "our existing customer base is growing revenue." NPS is a proxy for "customers feel good about us." Churn rate is a proxy for "customers keep paying."
Then you tie compensation, career advancement, or organizational status to the metric. Teams now have incentive to move the number.
Some of that movement comes from genuinely improving the underlying thing. Some comes from optimizing the measurement itself. Over time, the ratio shifts toward measurement optimization because it is cheaper and faster than real improvement.
Eventually the metric moves without the underlying reality changing. It has ceased to measure what it was supposed to.
Common cases in SaaS retention
1. NRR gaming
NRR targets are set. CS teams focus intensely on expanding the top 5% of accounts. Massive expansion in a few accounts pushes NRR above 110%. The base of smaller accounts, meanwhile, is churning at 8% quietly.
Reported NRR: 118%. Actual customer base health: declining. Two years later, the top-5% accounts consolidate their vendors and the growth story disappears.
2. NPS survey timing
NPS is a KPI. Someone realizes that surveys sent immediately after a positive product interaction get higher scores than surveys sent at random times.
Suddenly NPS is only sent after successful actions. Score improves from 32 to 51 over 6 months. Actual customer sentiment has not changed. The survey has just been re-timed.
3. Churn rate reclassification
Churn rate is being watched by the board. A cancellation this month would push the number above target. Someone convinces the customer to "pause" instead. The cancellation does not count as churn this quarter. It counts next quarter. Or it gets forgotten if the customer never comes back.
Reported churn: 3%. Actual customer loss: 4.5%. The definition of "churn" quietly expanded to exclude paused accounts.
4. Health score inflation
The CS team is measured on the percentage of accounts in "green" health status. Someone tweaks the health score rules to be more generous. More accounts turn green. CS performance appears to improve.
Reported healthy accounts: 78%. Actual account health: same as before. Just measured differently.
5. "Active user" definition expansion
Monthly active users is a growth metric. Someone expands the "active" definition from "logged in this month" to "logged in in the last 60 days." The number goes up.
Reported MAU: 45,000. Actual usage: unchanged. The window just doubled.
Signs your retention metrics are being gamed
1. Numbers are suspiciously clean
Real retention numbers wobble. If your NRR sits at exactly 108% for three quarters in a row, something is being managed. Real numbers oscillate; targeted numbers stabilize suspiciously.
2. Improvements coincide with metric definitions changing
A metric definition changes and the number improves in the same quarter. This is usually gaming, not improvement.
3. Uncorrelated metrics start correlating
NRR is improving. But gross retention is not. Product usage is not. NPS is not. Support tickets are not. If NRR is going up in isolation, it is probably being optimized at the expense of things it should correlate with.
4. Team behavior focuses on measurement, not customers
You start hearing: "How do we count this?" "Where does this show up in the metric?" "What if we redefined X?" These are the sounds of Goodhart in action. Teams are optimizing the measurement, not the outcome.
How to defend against Goodhart
Measure multiple correlated metrics
If NRR is a KPI, also track gross retention, product engagement, NPS, and support ticket volume. When they diverge, something is off. Multiple metrics that all should move together are harder to game than one.
Audit for gaming patterns
Regularly review whether metric definitions have shifted. Look for suspicious cleanness. Ask "what would the metric look like under the old definition?" and compare.
Rotate the KPI focus
If NRR is the KPI for two years and every team is optimizing it, switch the KPI to gross retention for a year. This forces broader retention work and prevents extreme optimization of any single metric.
Include qualitative context in reporting
Numbers alone hide gaming. Numbers with context (which accounts drove the change, what changed operationally, what customer feedback correlates) surface gaming.
Ensure metric owners do not control definitions
The person measured on NRR should not be the person who defines what counts as expansion revenue. That is a governance separation that most SaaS orgs skip.
The uncomfortable implication
Every metric in your retention dashboard is potentially subject to Goodhart. The better it looks, the more you should ask what is being gamed.
This is not paranoia. It is realism about how organizations respond to incentives. The teams with the healthiest retention metrics are not always the ones with the healthiest retention. Sometimes they are just the ones optimizing measurement most aggressively.
Real retention improvement is slower, messier, and harder to point at than metric gaming. That is why gaming is so common. And that is why healthy skepticism about your own numbers is a retention superpower.
Related concepts
- Survivorship bias in cohort charts - related measurement problem
- What is NPS? - the metric most vulnerable to gaming
- Net revenue retention - properly defined
To score whether your retention setup measures what you think it measures, take the 60-second Churn Health Check.