**What the metric is:** NPS asks customers "how likely are you to recommend us to a friend or colleague?" on a 0-10 scale. The score is the percentage of respondents who answered 9-10 (promoters) minus the percentage who answered 0-6 (detractors). Respondents answering 7-8 (passives) are excluded from the math. **What it's trying to measure:** Customer satisfaction sturdy enough to produce voluntary referrals. The intent is to capture not just "do customers like us" but "would they put their own reputation on the line for us." **The gap between the metric and the thing:** NPS has several known gaps. First, the question asks about an intention (would recommend) rather than a behavior (did recommend), and intention is a weak predictor of behavior in many contexts. Second, the math throws away the 7-8 responses, which means a respondent moving from 8 to 9 changes the score dramatically while a respondent moving from 6 to 8 doesn't. Third, cultural baselines vary — the same satisfaction in a US sample produces higher numerical scores than in a Japanese sample. The metric and the thing diverge whenever any of these conditions don't hold. **What "good" looks like:** Industry benchmarks vary widely by category. For SaaS specifically, scores above 30 are generally considered solid, above 50 strong, above 70 exceptional. But the absolute number matters less than the trend in your own data; a company moving from 12 to 24 is improving even if 24 is below the SaaS benchmark. **What "bad" looks like:** Several patterns matter beyond the level. A score that's flat for a year while you've shipped meaningful improvements probably means the survey isn't reaching the right people or the right moments. A score that's high but trending down is a leading indicator of churn. A score that's volatile by month usually means low sample size, and you should aggregate to quarters before reading anything into it. **How this metric gets gamed:** - Asking only happy customers. Surveying right after a positive support interaction inflates the number. - Asking right after a feature launch. Recency bias produces a spike that doesn't reflect ongoing experience. - Tying compensation to NPS. Customer-facing teams start coaching customers on how to answer ("give us a 9 if everything's been great"). - Filtering out specific cohorts. Removing power users, free users, or any specific segment shifts the score. **What to pair it with:** - Actual referral rate (do referrals happen, not just whether they're claimed to be likely). - Customer effort score or CSAT for the question of "did this transaction go well," which NPS doesn't capture. - Churn rate and expansion rate, which are the behaviors NPS is trying to predict. **Bottom line:** NPS is useful as a trend tracker for customer relationship health, and useless as a precise comparative benchmark. Trust the movement in your own number; don't trust comparisons against industry averages or competitor scores.