What Is a Good NPS Score? Compare Benchmarks and Survey Conditions
There is no NPS passing score that applies to every business. Even within one industry, country, language, customer population, timing, and survey design can change what a comparison means. Before adopting a benchmark, examine its conditions and the counts behind the score.
1. Avoid a universal definition of a good score
NPS summarizes recommendation intent. A positive score means the promoter proportion exceeds the detractor proportion; it does not by itself demonstrate a sufficient customer experience or higher profit. A negative score calls for investigation, but causes and urgency still depend on the evidence.
Do not turn an unsourced threshold such as “30 is good” or “70 is world class” into a universal target. If the reference study's methods are unavailable, use comparable internal history and customer explanations while leaving the external ranking undecided.
2. Include passives in the denominator
Bain's official explanation classifies responses of nine or ten as promoters, seven or eight as passives, and zero through six as detractors. Subtract the detractor percentage from the promoter percentage. Passives do not enter the difference in the numerator, but they remain in the denominator of valid respondents.
With 50 promoters, 30 passives, and 20 detractors among 100 valid respondents, NPS is 50% minus 20%, or 30. The denominator is not the 70 promoters and detractors alone. Keep nonresponse and invalid answers separate, with documented exclusion rules.
3. Check the benchmark's population and method
Confirm the year and region, then investigate who was recruited and how. Existing customers, recent users, and participants in an independent competitive study are not automatically equivalent populations. Copying only an industry average removes that context.
Condition | What to check | If conditions differ |
|---|---|---|
Country and language | The same region and question language | Do not apply an invented cultural correction |
Product and customer | Use case, customer type, and tenure | Identify the different population |
Scope of the question | Overall relationship or a specific experience | Do not rank unlike measures directly |
Recruitment and response | Invitation, mode, and response rate | Retain potential selection bias |
Timing | Year, season, and recent events | Avoid claiming a current ranking |
Calculation | Sample size, weights, and missing data | Do not pool unknown methods |
A general statement about scores being lower in a country does not establish an adjustment for a particular survey. Recognizing a comparability problem and calculating a defensible correction are different tasks. Record unknown conditions explicitly.
4. Separate relationship and episode measures
Bain's benchmark overview distinguishes overall and customer-episode levels. A question about the entire business relationship has a different reference point from a question asked after a particular service interaction.
For example, a fictional company could ask established customers about the relationship every six months and ask recent support users about their interaction. Placing those scores side by side would not prove one team performs better. Preserve separate series with clear populations and questions.
For repeated measurement, align wording, options, eligibility, and collection method. If a change is needed, mark it rather than presenting the entire history as unchanged. AAPOR's best practices are a useful reference for transparent survey reporting.
5. Pool counts carefully: a fictional example
This invented example assumes two compatible groups with the same question, eligibility, period, and no weighting. It is not an industry benchmark.
Group | Promoters | Passives | Detractors | Valid responses | NPS |
|---|---|---|---|---|---|
A | 50 | 30 | 20 | 100 | 30 |
B | 8 | 8 | 4 | 20 | 20 |
Pooled | 58 | 38 | 24 | 120 | Approximately 28.3 |
A is (50 − 20)/100 × 100 = 30. B is (8 − 4)/20 × 100 = 20. Pooling the counts gives (58 − 24)/120 × 100, approximately 28.3, rather than the simple score average of 25. The difference between A and B is ten score points, not a ten-percent increase. Apply a consistent rounding convention.
Pooling is appropriate only if the combined population and survey methods justify it. Different studies or weighting schemes cannot necessarily be combined using respondent counts alone. Small groups can also move substantially when a few answers change.
In B, changing one passive response into a promoter response moves NPS from 20 to 25: five score points. The equivalent change among 100 valid respondents moves the score by one point. Retain counts and response composition alongside changes. This arithmetic example is not a substitute for an uncertainty analysis.
6. Choose improvements from reasons and problems
Read why people selected their score. Slow response, confusing steps, and a mismatch with expectations require different actions. Respondents do not automatically represent every customer; investigate nonresponse or discontinued use where relevant.
After a change, repeat measurement under comparable conditions and report scores, counts, reasons, and observed behavior separately. Promoters are not a count of people who actually referred someone. Use the score to investigate and improve a concrete customer problem rather than optimizing the appearance of the number.
Related reading