By 
The maths behind an NPS calculator takes about ten seconds. Two subtractions and a division, and any spreadsheet will do it. Reading the number is the part that takes judgement, because an NPS of 40 can mean three completely different things about your customers, and on a small sample it might not mean 40 at all. So here is the calculation in full, a worked example you can copy, and then what the number is actually telling you: what counts as good, how many responses you need before it means anything, and what to do with it once you have it.
The NPS calculation, in one line
Net Promoter Score comes from a single question: “How likely are you to recommend us to a friend or colleague, on a scale of 0 to 10?” You sort the answers into three groups, and only two of them count.
Promoters score 9 or 10. Passives score 7 or 8. Detractors score anything from 0 to 6.
The formula is the percentage of promoters minus the percentage of detractors:
NPS = % Promoters − % Detractors
Passives are not ignored, which trips people up. They count towards your total response number, which drags both percentages down, but they are neither added nor subtracted. The result is a whole number between -100, where everyone is a detractor, and +100, where everyone is a promoter. It is not a percentage, and writing it as “40%” is the fastest way to signal you have not actually run the sum.
This grouping, and the whole metric, comes from Fred Reichheld’s 2003 Harvard Business Review article “The One Number You Need to Grow”, and the definitions are maintained by Bain on the Net Promoter System site. Every calculator you will find is running this same line of arithmetic.
A worked example you can copy
Say you send the NPS question to your last 200 buyers and every one of them replies. The scores break down like this:
| Group | Score | Count | Share of total |
|---|---|---|---|
| Promoters | 9 to 10 | 110 | 55% |
| Passives | 7 to 8 | 60 | 30% |
| Detractors | 0 to 6 | 30 | 15% |
| Total | 200 | 100% |
Promoters are 110 of 200, which is 55 per cent. Detractors are 30 of 200, which is 15 per cent. Subtract one from the other:
55 − 15 = an NPS of 40.
The 60 passives never entered the subtraction, but they are the reason your promoter share is 55 and not higher. That is the whole calculation. To do it on your own data, count your three groups, divide the promoter count and the detractor count each by your total responses, multiply by 100, and subtract. A spreadsheet with three cells does it. So does the back of an envelope.
The same score, three different businesses
Before you take that 40 anywhere, notice what it hides. The subtraction discards the passives entirely, which means very different customer bases collapse to the same number.
| Promoters | Passives | Detractors | NPS |
|---|---|---|---|
| 55% | 30% | 15% | 40 |
| 40% | 60% | 0% | 40 |
| 70% | 0% | 30% | 40 |
All three score 40. The first is an ordinary business with a solid core and a fixable tail. The second has almost nobody who dislikes it and almost nobody who loves it, which is a positioning problem rather than a service one. The third is polarising: a large group of advocates and an equally large group actively telling people not to buy. Those three situations call for completely different responses, and the score on its own cannot tell them apart. Always report the three shares underneath it, not just the number they produce.
What counts as a good score
This is the question the calculator can’t answer, and where the real judgement sits. The ladder you will see quoted everywhere runs: above 0 is good, above 20 is favourable, above 50 is excellent, above 80 is world class. It is almost always credited to Bain, and that credit is misplaced. Bain’s own page on measuring your score publishes no absolute thresholds at all. The only benchmark it offers is relative: that sustained value creators run an NPS around twice the average company’s. The familiar ladder is Qualtrics’ framing, and it is a perfectly reasonable rule of thumb as long as you know whose rule of thumb it is. Published cross-industry figures cluster somewhere in the low 30s to mid 40s depending on whose pool you use, so most companies land in the favourable band and short of excellent.
Two warnings before you compare yourself to any of that. First, benchmark tables vary a lot between the providers who publish them, because each is drawn from a different pool of customers, so treat any single industry figure as directional rather than a target. Second, and more useful, the comparison that tells you something is your own score over time. An NPS of 30 that was 20 last quarter is a better story than an NPS of 45 that was 55. A global median cannot tell you whether you are getting better. Your own trend can.
How many responses you actually need
NPS takes an 11-point scale and crushes it into three buckets, and that throws away information. Jeff Sauro’s team at MeasuringU and the analysts at Displayr have both shown that this collapsing roughly doubles the margin of error around the score compared with just averaging the raw ratings. In plain terms, the number is jumpier than it looks.
What that means day to day: below roughly 100 responses, your NPS can move by double digits from one batch to the next on pure chance, with nothing about your business having changed. A company quoting an NPS of 50 might genuinely be sitting anywhere from the mid-30s to the mid-60s once you account for the error. And to prove that a change over time is real rather than noise, you need a bigger sample than most people expect, often around twice the size you would guess.
You can check this on your own numbers instead of taking my word for it. Treat every response as a value: +1 for a promoter, 0 for a passive, −1 for a detractor. Your NPS is just the average of those values written as a whole number, and the uncertainty around it is the standard deviation of that set divided by the square root of your sample size, multiplied by 1.96 for a 95 per cent interval.
Run it on the worked example above. With 55 per cent promoters, 30 per cent passives and 15 per cent detractors, the standard deviation works out at 0.73, so at 200 responses the margin of error is about 10 points. That tidy NPS of 40 is really somewhere between 30 and 50. At 100 responses the same spread gives you roughly plus or minus 14. At 400 it tightens to about 7. Halving the error costs four times the responses, which is the uncomfortable arithmetic sitting underneath every NPS dashboard.
None of this makes NPS useless. It makes it a metric you read like a trend line, not a thermometer. Collect enough responses that a normal week’s variation does not swamp the signal, watch the direction over several periods, and resist the urge to explain every wobble in the third decimal place. If you get 40 responses a month, do not report a monthly NPS. Report a quarterly one.
Transactional or relational, and the question wording
There are two ways to run the survey, and they answer different questions.
Relational NPS asks about the whole relationship on a schedule, say once a quarter, and gives you a broad read on customer health. Transactional NPS fires straight after a specific moment, a purchase, a delivery, an onboarding call, a support ticket, and tells you whether that moment landed. In ecommerce that usually means putting it in a post-purchase survey. Transactional surveys usually get higher response rates because the experience is fresh and the question feels relevant, where a relational survey asks people to average their whole history with you in their head. Most mature programmes run both: relational for the trend, transactional to find the moments that make or break it.
For response rates, hold your expectations sensibly. Email NPS surveys commonly land somewhere in the 10 to 30 per cent range, with in-app and post-purchase placements often doing better, according to SurveyMonkey’s benchmark data. If you want a reliable score, that response rate is doing as much work as the calculator is, because it decides your sample size. Most customer feedback tools will run the survey and track the trend for you.
Keep the question itself standard. The 0 to 10 scale, the word “recommend”, and a follow-up asking why. If you are building the rest of the survey around it, our Likert scale examples cover how to word the other rating questions. Changing the wording or the scale makes your score incomparable to everyone else’s and, more importantly, to your own past score.
The number is the scoreboard. The reasons are the game.
The single biggest mistake with NPS is treating the number as the output. It is not. The number tells you which direction you are pointing. The open follow-up, “what is the main reason for your score?”, tells you what to do about it, and that is the part worth reading every week.
A detractor who writes “delivery took nine days” is a fixable operations problem. A detractor who writes “it did not do what the ad implied” is a positioning problem that will keep costing you refunds. Both count the same in the calculation and mean completely different things to your business. Coding those written answers into a handful of recurring themes, then counting the themes, is where NPS stops being a vanity metric and starts driving decisions. It is also worth writing the follow-up properly, since a weak open question returns weak verbatims; there are worked examples of open-ended questions that earn a real answer. If you are setting up to listen to customers at scale, it is worth comparing the voice-of-customer tools built for exactly that job, because analysing the verbatims is where the return is.
Where NPS runs out of road
NPS has one structural limit worth naming. It only measures the people who already bought from you and stuck around long enough to answer. It is a rear-view mirror. That is genuinely valuable for improving what you already sell, but it tells you nothing about the customers you have not won yet, or about whether a new product, pack or price will land before you commit money to it.
It is also worth knowing that the metric’s headline claim is contested. Reichheld’s original argument was that NPS predicts growth better than other measures. When Keiningham and colleagues tried to replicate that across 21 industries for the Journal of Marketing in 2007, they found no support for it: NPS predicted growth about as well as a conventional satisfaction score, and no better. That does not make it worthless. A single number the whole business understands has real organisational value, and the follow-up question earns its keep on its own. It does mean treating NPS as one useful input on customer sentiment rather than, as the original title had it, the one number you need to grow.
That forward-looking job is a different one, and it belongs before you spend rather than after. It means testing the idea against the audience you are targeting ahead of building or launching it. That is the one place our own tool fits, so I will name it plainly. With TestFeed you can put a product, a pack, an in-context price, an ad or a claim in front of shoppers modelled on your target customer and get back a purchase-intent read, their reasons in their own words, and a clear next move, in days rather than weeks. It is a directional pre-launch signal, not a sales forecast, and it does not judge taste, texture or smell. Use it to decide what is worth launching. Then point an NPS survey at the customers it brings you, and read the trend.
Frequently asked questions
How do you calculate NPS?
Ask customers how likely they are to recommend you on a 0 to 10 scale. Count promoters (9 to 10) and detractors (0 to 6). Turn each into a percentage of your total responses, then subtract the detractor percentage from the promoter percentage. If 55 per cent are promoters and 15 per cent are detractors, your NPS is 40. Passives, who score 7 or 8, count towards the total but are not added or subtracted.
What is a good NPS score?
The ladder usually quoted is above 0 good, above 20 favourable, above 50 excellent and above 80 world class. It is widely credited to Bain, but Bain publishes no absolute thresholds; that framing is Qualtrics’. Published cross-industry figures cluster in the low 30s to mid 40s depending on the pool. A good score depends on your industry and your customers anyway, so the more useful comparison is your own score last quarter, not a global average.
How many responses do you need for NPS to be reliable?
More than most people collect. Because NPS collapses an 11-point scale into three groups, the margin of error is wide. Below roughly 100 responses the number can move by double digits on sampling noise alone, and detecting a genuine change over time takes a larger sample still. Treat a small-sample score as a rough read and follow the trend across several periods rather than reacting to a single figure.
Is NPS a percentage?
No. Although it is built from two percentages, NPS is reported as a whole number between -100 and +100. An NPS of 40 is not 40 per cent. Writing it with a per cent sign is one of the most common mistakes and a quick sign the person quoting it has not calculated it themselves.
What is the difference between transactional and relational NPS?
Relational NPS asks about the overall relationship, usually on a schedule such as quarterly, and gives you a broad health read. Transactional NPS fires straight after a specific event, a purchase, a delivery, a support ticket, and tells you whether that moment worked. Transactional surveys tend to get higher response rates because the experience is fresh and the question feels relevant. Most programmes run both.
