How the NPS score is built
Net Promoter Score comes from one question (how likely are you to recommend us, on a scale of 0 to 10) and one piece of arithmetic:
| Answer | Group | Counts as |
|---|---|---|
| 9 to 10 | Promoters | +1 |
| 7 to 8 | Passives | 0 |
| 0 to 6 | Detractors | −1 |
NPS is the percentage of promoters minus the percentage of detractors, shown as a number from −100 to +100. Passives count in the denominator but add nothing to the result. That is the first thing to understand about it.
The number nobody reports: the margin of error
This is what sets this calculator apart from most others. NPS is an estimate from a sample. Unless you surveyed every customer you have, the score you calculated isn't the score of your customer base. It's one draw from a distribution.
You can calculate how wide that distribution is. Promoters and detractors come from the same sample and can't overlap, so the variance of the difference is:
Var(NPS) = [ p_prom + p_det − (p_prom − p_det)² ] ÷ n
The subtracted term is the part naive calculations miss. Treating the two proportions as independent samples would overstate the error, because a respondent who is a promoter can't also be a detractor.
| Responses | Margin at 95% | What that means |
|---|---|---|
| 50 | ≈ ±20 points | You can tell almost nothing apart |
| 100 | ≈ ±14 points | Only very large moves are real |
| 400 | ≈ ±7 points | Quarterly trends become readable |
| 800 | ≈ ±5 points | Month-to-month movement means something |
The consequence is uncomfortable, so here it is plainly: a company reporting "our NPS went from 31 to 42 this quarter" on 120 responses has measured a change that is statistically indistinguishable from zero. The slide deck says the score improved. The statistics say nobody knows. Don't reorganize a team or cut a budget on a move like that.
Why the margin of error shrinks so slowly
Error falls with the square root of sample size, not with sample size itself. Going from 100 to 400 responses, four times the survey effort, cuts the margin in half. Halving it again takes 1,600.
That's why chasing finer and finer precision on NPS stops paying off after a point, and why the smarter move is usually the opposite one: accept a wider interval, survey less often and compare against a longer baseline instead of last month.
What the NPS number does not tell you
- Improvement can be invisible. Moving a customer from 4 to 7 is a real fix to a real problem, and NPS doesn't move at all, since both count the same as any other non-promoter. A team chasing the headline number has every incentive to ignore that customer entirely.
- Two very different companies can score the same. 60% promoters and 20% detractors gives +40. So does 40% promoters and 0% detractors. The first has an active problem with a fifth of its customers, and the second doesn't. The distribution matters more than the score.
- The people who answer aren't the people you serve. Survey response rates run in the single digits, and the ones who respond tend to be the ones with a strong opinion. Non-response bias is usually bigger than the sampling error the interval measures, and unlike sampling error, it doesn't shrink with more responses.
- It doesn't predict much by itself. The original claim tied NPS to growth. Later replication has been mixed, and the relationship varies by industry. Treat it as one signal among several, not as a target to manage toward.
Comparing NPS scores, and when not to
Cross-company comparison is where NPS gets abused most. Two built-in distortions make most published comparisons meaningless.
Culture changes how people use a scale. Respondents in some countries avoid the extremes, so a truly satisfied customer answers 8 instead of 10, lands in the passive bucket and adds nothing. Scores from Japan and the Netherlands run structurally lower than scores from the United States for the same underlying satisfaction. Put a global average next to a US benchmark and you're comparing two different measurements.
Method changes the sample. An in-app prompt shown right after someone finishes a task successfully will produce a much higher score than an email survey sent to the full customer list, including people who churned. Neither is wrong. They measure different populations. And when a vendor publishes an industry benchmark, it rarely says which method sits behind it.
The comparison that survives both problems is your own score against your own score, collected the same way, with the interval shown.
Getting more out of the same NPS survey
The score is the least informative thing the survey produces. Three other outputs are worth more to you:
- The full 0 to 10 distribution. A two-humped shape, with a cluster at 9 to 10 and another at 0 to 2, is a completely different business situation from everyone bunched at 7 to 8, and both can produce the same NPS.
- The follow-up question. "What is the main reason for your score?" is where the information you can act on lives. The number tells you something changed; the text tells you what to fix.
- Segments. NPS by plan, tenure or acquisition channel usually shows that the overall score is an average of two very different populations. Segment carefully, though: splitting 200 responses into four segments leaves each with a margin near ±28 points.
Privacy
Everything runs in this tab as arithmetic. Nothing is uploaded, nothing is logged and there is no account. That matters, because response counts and satisfaction data are usually internal figures.
Frequently asked questions
Why does this calculator show a confidence interval?
NPS is an estimate from a sample, not a measurement, and reporting it as one exact number is how teams end up making decisions on noise. With 100 responses the 95% margin is roughly ±14 points. A team that reports moving from 31 to 42 on that sample size has measured nothing you can tell apart from no change at all. The interval puts that in plain view instead of hiding it.
How many responses do I need?
That depends on your answer mix, which is why the tool calculates it for you. As a rough guide, getting the margin down to ±5 points usually takes 700 to 900 responses, and ±10 points takes around 200. Below 100 responses, the margin is usually wider than any month-to-month movement you'd want to act on.
Why are 7 and 8 excluded from the score?
By definition, passives count in the denominator but add nothing to the numerator. The split is a convention Fred Reichheld introduced in 2003, not something derived from statistics. In practice, a company can get better, moving detractors up to 7, and see no change in NPS at all. That's a known limitation, and a reason to look at the full distribution instead of the headline number alone.
Can I compare my NPS to another company's?
Carefully. Scores don't compare across countries, because people use rating scales differently in different places. Respondents in some cultures avoid the ends of a scale, which pushes NPS down regardless of how satisfied they actually are. Scores don't compare across survey methods either: an in-app prompt after a successful action produces a very different sample from an email to your whole list. The most reliable comparison is your own score against itself, measured the same way over time.
Is my data sent anywhere?
No. Every calculation runs in your browser as plain arithmetic. Nothing is uploaded, nothing is stored and there is no account.