NPS calculator

The NPS formula is a subtraction any spreadsheet can do. What this calculator adds is the uncertainty around the number. NPS is an estimate drawn from a sample, not a fact about your whole customer base, and every estimate carries statistical slack. Ignoring that slack is where the monthly review meeting comes from, the one where somebody explains why the score dropped four points when it did not drop at all. Here you get the score, the margin of error, the real range your NPS sits in, the arithmetic path to your goal and an honest test between two periods.

How to use it

1. Pick the mode: NPS of one survey (the default) or compare two periods.
2. Enter how many responses landed in each bucket: promoters (9 and 10), passives (7 and 8) and detractors (0 to 6). Count only valid responses, not people who dropped out.
3. Set the confidence. Use 95%, the market standard. Move to 99% when the decision is expensive and hard to undo.
4. Choose the benchmark industry for context and the NPS goal you are chasing.
5. Read the range before the score. If the distance to your goal is smaller than the margin of error, the tool says so: that sample cannot tell whether you hit it.

How it works: the formula

The metric itself is a difference of proportions, and the margin of error comes from the variance of that difference inside the same sample:

Where p is the promoter share, d is the detractor share, n is the total number of responses and z is the normal critical value for the chosen confidence (1.96 at 95%). The detail most tools get wrong is the second line: promoters and detractors come from the same survey and are negatively correlated, because a response that becomes a promoter can no longer be a detractor. Treating the two proportions as independent inflates the uncertainty for no reason. The fourth line is what makes the period comparison valid: since the two rounds are separate samples, the variances add up, and the test is an ordinary two sided z.

Worked example (reproduces the default output)

With the prefilled values: 420 promoters, 330 passives and 250 detractors, on 1,000 responses. The shares are p = 0.42 and d = 0.25, so NPS is (0.42 − 0.25) × 100 = 17.0 points. The variance is (0.42 + 0.25 − 0.17²) / 1,000 = (0.67 − 0.0289) / 1,000 = 0.0006411. Its square root is 0.025320, which is 2.53 points of standard error. At 95% confidence the margin is 1.96 × 2.53 = 4.96, which the tool rounds to 5.0 points, and the real range runs from 12.0 to 22.0. In other words: your NPS is not 17, it is somewhere between 12 and 22.

The default goal of 30 points sits 13 points away. Since each converted passive adds 100/1,000 = 0.1 point, you need 13 / 0.1 = 130 passives turning into promoters. Down the detractor path each conversion adds 200/1,000 = 0.2 point, so 65 detractors are enough. Same goal at half the effort, which is why recovering unhappy customers moves the metric faster than delighting the merely satisfied.

In compare mode the defaults hold 420/330/250 for the previous period (NPS 17.0) against 455/320/225 for the current one (NPS 23.0). The gap is 6 points, the standard errors are 2.53 and 2.50, and the error of the difference is √(2.53² + 2.50²) = 3.56. The margin at 95% lands on 1.96 × 3.56 = 7.0 points, the interval runs from minus 1.0 to 13.0 and the p-value is 9.2%. Translated: NPS went up 6 points on the chart and the data cannot say it went up at all.

Margin of error by sample size

The table below uses the same mix as the example (42% promoters, 33% passives, 25% detractors) at 95% confidence. It is the most useful argument on this page for a board discussion: it shows which swing deserves attention and which one is noise.

Notice the square root at work: quadrupling the sample halves the margin. Going from 500 to 1,000 responses buys 2 points of precision, while going from 5,000 to 10,000 buys less than one. There is a point where collecting more responses stops paying for itself, and it arrives much earlier than most teams assume.

How to read it and where NPS misleads

The first limit is the one the margin already exposes: NPS from a small survey does not support monthly tracking. At 100 responses a month the real range covers 30 points of the scale, so any time series turns into random jitter. Either you pool responses into wider windows, or you agree to only compare large moves.

The second is the information the rule throws away. A 7 and a 0 are treated very differently, yet a 0 and a 6 are the same thing to the metric, exactly like a 9 and a 10. That means a base can migrate from zeros to sixes, a huge real world improvement, without moving NPS a single point. Always read the full score distribution next to the metric, never the headline number alone.

The third is cross company comparison. There is documented cultural bias in how people use a 0 to 10 scale, on top of differences in when the question is asked (right after a successful purchase or in the middle of the usage cycle), who gets invited and which channel was used. Comparing your NPS against a number published by another company almost always compares methodologies, not experiences. The comparison that holds is against yourself over time, with the method kept fixed.

The fourth is the costliest: NPS does not tell you what to do. It compresses satisfaction into a scalar, and a scalar carries no cause. Nearly all of the value of the survey lives in the open text field after the score and in reading it by theme. Companies that turn the number into a bonus target usually get biased collection, with staff asking for high scores instead of fixing problems, and the metric stops measuring what it was meant to measure.

Measurement best practices

• Freeze the method: same question, same channel, same moment in the lifecycle, same invitation rule. Change the method and you break the series.
• Publish the margin of error next to the score. A metric without uncertainty becomes a narrative.
• Keep transactional NPS (right after an interaction) separate from relational NPS (whole base, periodic). They are different measures and should not be pooled.
• Track the response rate too. A drop in participation changes who answers and moves the score on its own.
• Prioritise detractors by volume of reason, not by score. The text next to the number is worth more than the scalar.
• Before announcing an improvement, run the period comparison. If the interval crosses zero, you have noise, not a result.

FAQ

How is NPS calculated?

NPS is the percentage of promoters (scores 9 and 10) minus the percentage of detractors (scores 0 to 6). Passives (scores 7 and 8) count in the denominator but score nothing. With 420 promoters, 330 passives and 250 detractors out of 1,000 responses, the math is 42% minus 25%, which gives 17 points. The result is always stated in points from minus 100 to plus 100, never as a percentage.

What is a good NPS?

The classic reading is: above zero means a positive balance, above 30 is considered great and above 70 is excellence. The generic scale is misleading on its own, though, because the normal level swings a lot by industry: telecom lives near 25 while retail and travel run past 45. That is why this tool compares your number against the median of the industry you pick, not only against the generic bands.

What is the margin of error of NPS?

It depends on sample size and on how the answers are spread. With 1,000 responses in the example mix, the margin at 95% confidence is roughly 5 points. With 100 responses it climbs to about 15.7 points, and with 10,000 it drops to about 1.6 points. That is why a company running a small survey should not chase monthly swings: the whole movement fits inside the margin.

My NPS went up 6 points. Is that real?

Only the test can answer. In compare mode, with 1,000 responses in each period, a move from 17 to 23 points comes with a 7.0 point margin and a p-value of 9.2%: the difference fits inside sampling noise, so the data does not support the celebration. With larger samples that same 6 point gap would become significant. What changes is not the size of the improvement, it is your ability to see it.

How many responses do I need to measure NPS?

Pick by the margin you can live with. If a 5 point difference changes your decision, you need something close to 1,000 responses per round. If 10 points is enough, about 250 will do. Below 100 responses NPS is close to decorative, because the margin passes 15 points and covers half of the useful scale. To plan the send, use the survey sample size calculator.

Is it better to convert a passive or a detractor?

A detractor, always, and the reason is arithmetic: moving one response from passive to promoter creates a promoter and adds 100/n points, while moving one from detractor to promoter creates a promoter AND removes a detractor, adding 200/n points. Every converted detractor is worth exactly double. Detractors also tend to state a concrete reason, which makes them easier to act on than passives, who rarely articulate what was missing.

Keep going

Still about to send the survey and wondering how many responses to collect? Start with the survey sample size calculator. If your problem is the uncertainty range of a conversion rate instead of a score, use the confidence interval calculator. And if the plan is to turn dissatisfaction into tested improvements, follow the conversion rate optimization guide alongside the conversion rate calculator.