NPS earned its popularity honestly: one question, one number, easy to track over time, easy to report to a board. Those same properties make it a poor thing to manage a team against, and the failure mode is predictable enough that it's worth naming directly.
The number has no diagnostic power by itself
An NPS of 32 tells you almost nothing actionable on its own. It doesn't tell you which part of the experience is driving detractors, whether the score moved because of a real change or a sampling shift, or whether your promoters are promoters because of the product or because of a discount they just received. A single compressed number can't carry that much information, no matter how precisely it's calculated.
Measurement artifacts move the score more than reality does
Surveying customers immediately after a support interaction produces different scores than surveying them at a random interval, because who gets surveyed and when both quietly bias the sample. A shift from in-app surveys to email surveys changes the respondent population, which changes the score, independent of any actual change in sentiment. Teams that track NPS quarter over quarter without controlling for these factors are frequently reacting to measurement noise, not signal.
Managing to the number changes what gets optimized
Once NPS becomes a target with consequences attached, the fastest way to move it is rarely to improve the underlying experience — it's to change who gets asked, when they get asked, or how the question is framed. This is a specific case of a much more general pattern: any metric that becomes a target stops being a reliable measure, because people (reasonably) start managing to the metric rather than the outcome it was meant to represent.
What to do instead
- Keep NPS as a comparative benchmark, tracked consistently with fixed timing and channel, useful for spotting large directional shifts — not as a target with consequences attached
- Segment respondents by tenure, plan tier, or use case before looking at the score — an aggregate NPS can hide a serious problem in one segment entirely
- Treat the verbatim comment as the real data, and code it into themes systematically, rather than skimming a handful of quotes for a board deck
- Pair NPS with a behavioral metric that isn't self-reported — renewal rate, expansion revenue, support ticket volume — since self-reported sentiment and actual behavior frequently diverge
NPS is a reasonable thermometer. It's a bad thermostat. Used to notice large directional shifts across a stable measurement process, it's useful. Used as something a team is compensated against, it reliably produces exactly the wrong incentives.