Net Promoter Score (NPS): Definition & Calculation

Two firms report the same Net Promoter Score. At the first, half the customers are delighted and half furious. At the second, nobody complains and nobody is impressed.

The Net Promoter Score measures how willing customers are to recommend you: you ask on a scale from 0 (very unlikely) to 10 (very likely), sort the answers into promoters, passives and detractors, then subtract the percentage of detractors from the percentage of promoters. The result lands between −100 and +100. At empirio.ai, an online survey tool from Germany, matrix, ranking and NPS questions are free. By the end of this article you will have worked out an NPS yourself, placed it and spotted where it promises more than it delivers.

📌 Key points at a glance

• The NPS measures how likely customers are to recommend you, on a scale from 0 to 10.
• Promoters answer 9 or 10, detractors answer 0 to 6.
• NPS = % promoters − % detractors, passives (neutrals) do not count.
• The score lies between −100 (negative) and +100 (positive).
• Fred Reichheld introduced the NPS in the Harvard Business Review in 2003.

What is the Net Promoter Score?

The Net Promoter Score, or NPS for short, is a customer loyalty metric that grows out of a single question:

“How likely is it that you would recommend us to a friend or a colleague?”

Answers run on a scale from 0 (very unlikely) to 10 (very likely).

Credit for the metric goes to the American loyalty researcher Fred Reichheld, who set it out in the Harvard Business Review in December 2003 under the title “The One Number You Need to Grow”. His argument was blunt: of all the questions you could put to a customer, one says more about future growth than any other, and that is the question about recommending you. Recommending a company means staking your own reputation on it, which is precisely why a recommendation counts for more than a tick in a box marked “satisfied”.

One detail most guides skip over is the legal one. “Net Promoter” and “NPS” are registered trademarks of Bain & Company, Inc., NICE Systems, Inc. and Fred Reichheld. Applying the method is free and publishing your own score is fine, but commercial use of the marks requires a licence and an attribution under Bain’s own terms on trademarks and licensing (as at September 2026). For a survey in a club, a course or your own shop that makes no difference at all. For an annual report it does.

Promoters, passives (neutrals) and detractors: the three groups

The eleven answer options (0 to 10) are later collapsed into three groups when you analyse the results:

• Promoters with 9 and 10: recommend you actively and stay for years.
• Passives (neutrals) with 7 and 8: content, but not attached.
• Detractors with 0 to 6: unhappy, and they tell people.

That split is the real heart of the method and it is lopsided, which throws most people on first contact. Seven of the eleven possible answers count as criticism. Reichheld justifies the cut with behaviour rather than symmetry: in his data, people who gave a 9 or a 10 behaved measurably differently from everybody else. They bought again more often and they brought new customers with them. The line falls where the behaviour changes, not in the middle of the scale.

A 0 to 10 scale is a rating scale with eleven steps. Laid out on screen it looks like a ruler with even gaps, and it is nothing of the sort: nobody can show that the step from 6 to 7 means the same as the step from 9 to 10. Which is exactly why the NPS counts shares instead of taking an average.

How to calculate the NPS: formula and worked example

The NPS is calculated by subtracting the percentage of detractors from the percentage of promoters. Passives drop out of the sum entirely and only count towards the total number of answers. Report the result as a whole number, without a per cent sign.

1. Sort the answers into three groups. Count how many answers fall on 9 to 10, on 7 to 8 and on 0 to 6.
2. Work out the shares in per cent. Divide each group by the total number of answers and multiply by 100.
3. Subtract detractors from promoters. What remains is your Net Promoter Score.

From the three group figures you can calculate the NPS straight away.

A worked example with 200 completed questionnaires from a customer survey makes the arithmetic concrete.

From 48 per cent promoters minus 25 per cent detractors you get an NPS of +23. The 54 passive answers change nothing directly, they only press the figure down indirectly by swelling the total and shrinking both shares. Had the same 96 promoters and 50 detractors come from only 150 answers, the NPS would be +31.

What counts as a good NPS?

No universal threshold for a good NPS exists. Everything depends on sector, country and how you collected the answers, which is exactly why measuring yourself against other people’s figures is the commonest mistake of all. The one safe statement is whether promoters or detractors have the upper hand. For a rough steer, our NPS calculator uses the bands below. Treat them as a reading aid, not a standard:

• Below 0: critical. Detractors outnumber promoters, so read the open answers before anything else.
• 0 to +29: room to improve. Promoters are level with or ahead of detractors, but with little to spare.
• +30 to +49: good. Clearly more advocates than critics.
• +50 to +69: very good. Most of the people you asked would recommend you of their own accord.
• +70 to +100: excellent. Rare in practice, which is all the more reason to show the shares and the number of answers next to it.

The dependable reference point is your own time series. Measure the NPS over several quarters with the same question, the same moment in the customer journey and the same audience, and the movement tells you something real. Bain & Company sells sector benchmarks under the name NPS Prism, and that market exists because a bare score means so little without context. For an outside view, Britain offers something better than a scattering of self-reported scores: the UK Customer Satisfaction Index (UKCSI) has been published twice a year by the Institute of Customer Service since 2008, always on a consistent set of measures.

Response behaviour also varies between countries, which international surveys underrate with some regularity. Van Herk, Poortinga and Verhallen (2004) reported in the Journal of Cross-Cultural Psychology that the tendency to agree and the pull towards extreme scores were stronger in the Mediterranean countries studied than in north-western Europe. The United Kingdom was one of the six countries in that study and sat in the north-western group. A British score and a Spanish or Italian score from the same company are therefore not directly comparable.

Asking the NPS question properly in a questionnaire

The NPS question only works when wording, scale and timing line up. Reword the question, shorten the scale or ask at some arbitrary point, and you are measuring something else, at which point the comparison with your own last reading falls apart.

Placement matters as much as wording. Put the question at the very end of the questionnaire or let it stand on its own. Ask it after ten detailed questions about delivery times and packaging and your respondents will answer with those details in mind rather than the whole picture. How you structure a survey helps decide what score comes out the other end.

Checklist for the NPS question

• The wording stays identical across every wave.
• The scale runs from 0 to 10, with nothing trimmed off.
• Both ends carry a label, such as “very unlikely” and “very likely”.
• An open follow-up question asking why sits directly underneath.
• The moment in the customer journey is the same every time.
• The question is voluntary, never a required field.

Running the NPS yourself: question type, plan and data protection

Running the NPS yourself takes three things: a question type that offers the full 0 to 10 scale without you rebuilding it, a plan whose free tier carries the analysis as well as the questionnaire and a lawful basis for processing the answers. Get those three right and the rest of the job is wording and timing.

The question type is the part most people underestimate. Rebuilt out of a rating scale or a drop-down list, an NPS question usually loses something along the way: the labels at both ends, the eleventh option or the grouping in the analysis. A dedicated NPS question type keeps the scale intact and sorts promoters, passives and detractors for you, so your figure does not depend on anyone remembering that 7 and 8 drop out of the sum.

Timing and permission belong to the same decision. An NPS survey mailed to a customer list is a different legal question from one sitting on a thank-you page behind an open link, and the lighter route often produces the better answers too, because nobody feels chased.

What the free plan should cover

A free plan is enough for an NPS survey as long as it carries the question type and the analysis, not just the questionnaire. The free plan at empirio.ai includes matrix, ranking and NPS questions, the analysis tools, the results as charts and the real-time updating, with 50 participants per month (as at October 2026), which is enough to create an online survey for free and read your first score the same week.

Two limits are worth knowing before you plan a wave. Cross-tabulations and filters start with the Plus plan, so splitting your NPS by customer segment is a paid step (as at October 2026). EU hosting is already in the free plan and so is the data processing agreement, which matters for the legal duty rather than for the arithmetic.

Data protection in NPS surveys: the duty first, then the solution

NPS answers are personal data in most set-ups, and the moment you hand them to an outside survey tool you are a controller using a processor. Article 28(3) UK GDPR requires that processing by a processor is governed by a contract, and the Information Commissioner's Office explains on “When is a contract needed and why is it important?” that a written contract is needed whenever a controller uses a processor, with the same requirement passed down to any sub-processor under Article 28(4). The duty sits with you as the controller, not with the tool you picked.

empirio.ai includes the data processing agreement in the free plan and hosts in the EU in the free plan as well (as at October 2026), so the contract is a document you accept rather than a negotiation you open. Keep a copy with your record of processing activities and note who accepted it and when.

Anonymity is a separate promise and a harder one to keep than a contract. An NPS survey is anonymous only when the link is open, no email address or account is tied to a response and the result is reported in groups large enough that nobody can be picked out of them. In a customer base of twelve, “we had three detractors” comes close to naming them, so publish shares only where the group is big enough to hide in.

Embedding the NPS question in your own website

Embedding beats sending for an NPS question, because the question reaches people where the experience just happened. Put the single question on the order confirmation page, in the help centre once a ticket closes or on a page customers visit anyway, and you hear from people who never had to open an email first. Embedding a survey is free at empirio.ai (as at October 2026), so the route costs nothing to try.

One rule survives the move into your own page: the question stays voluntary and the open follow-up question stays underneath it. An embedded NPS question that blocks the page or demands an answer before a customer can carry on produces compliance, not an opinion. To set one up and watch the score calculated as the answers arrive, you can create a free empirio.ai account and start with the NPS question type.

Advantages and disadvantages of the NPS

The NPS has spread through so many organisations largely because it is so simple. That same simplicity is what leaves it open to attack. Before you settle on the metric, it is worth weighing up both sides with a cool head.

The case for the NPS

• One question, answered in a matter of seconds, which keeps the response rate up.
• A figure between −100 and +100 needs no explaining, whichever department reads it.
• Keep the question and the timing the same and you build a trend you can rely on, quarter after quarter.
• The metric fits in plenty of places: after a purchase, after a support contact or once a year across your whole customer base.
• Paired with the open follow-up question, it gives you the reasons behind the score as well as the score itself.

The case against the NPS

• The distribution disappears: a divided customer base and an indifferent one can produce exactly the same score.
• Passives, the 7s and 8s, leave no trace in the result, even though they are often the largest group.
• With only a handful of responses the score lurches about although nothing has actually changed.
• What it captures is an intention, not a recommendation anyone has actually made.
• Whether the NPS predicts growth better than other metrics is still contested in the research.

What the NPS cannot do

The NPS is a deliberate simplification and every simplification carries a price. Anyone putting the score into a presentation should know three limits, because they explain how two identical scores can describe completely different situations.

None of these limits makes the NPS worthless. They mark where its remit ends. As an early warning system the metric earns its keep, as the sole basis for a decision it does not.

The NPS throws information away

Eleven gradations become three groups and three groups become one number. A company with 50 per cent promoters and 50 per cent detractors scores zero. A company with 20 per cent promoters, 60 per cent passives and 20 per cent detractors also scores zero. The first describes a split customer base with loud voices on both sides, the second a uniformly lukewarm one. Same number, two entirely different jobs to do.

Small samples make the NPS jump

The NPS is a difference between two shares, so it swings harder than an average taken from the same sample. At 50 answers a single response is worth 2 percentage points. If one person shifts from detractor to promoter, the score moves 4 points without anything at all having changed about the company. Report a 5 point improvement off 40 answers and you are reporting noise. How big a sample needs to be is no side issue here.

The link between NPS and growth is disputed

Reichheld’s central claim was that the NPS predicts growth better than rival loyalty measures. Keiningham, Cooil, Andreassen and Aksoy tested that in 2007 in the Journal of Marketing, using data from 21 companies and more than 15,500 interviews in the Norwegian Customer Satisfaction Barometer. Their finding: the claimed clear superiority of the NPS over other measures could not be confirmed in the very sectors Reichheld had held up as showcases. The paper won the 2007 H. Paul Root Award from the Marketing Science Institute.

The NPS is a thermometer, not a diagnosis. It tells you something is hot or cold, never why.

How the NPS differs from other metrics

The NPS looks ahead and stands for loyalty to your organisation as a whole. Three other metrics are often mentioned in the same breath. Like the NPS, they get by on a single question, but they measure something different:

• The CSAT (Customer Satisfaction Score) measures how satisfied someone was with one particular experience, such as a purchase or a call to customer services. It looks back.
• The CES (Customer Effort Score) measures how much effort it took to get something sorted. It suits processes such as placing an order or making a complaint.
• The eNPS (Employee Net Promoter Score) puts the recommendation question to your own staff. It shows whether employees would recommend their organisation as a place to work.

Which metric suits which situation, and where they overlap, is set out briefly in our comparison of NPS, CSAT and CES.

Common mistakes in NPS surveys

Most unusable NPS results come from the circumstances of the survey rather than from a slip in the arithmetic. Three patterns turn up again and again, and all three cost nothing to avoid.

Common to all three is that they push the score upwards. That is what makes them treacherous: an inflated NPS gets queried by nobody, because it is the pleasant news.

The NPS question arrives at the wrong moment

Straight after a complaint has been put right, people answer with relief and a generous hand. Three weeks later they answer far more soberly. Both readings are correct on their own terms, and together they build a time series that says nothing about the company and everything about when you chose to ask. Fix the moment once and hold to it, even when another moment looks more flattering.

The NPS becomes a target instead of a measurement

Once a bonus hangs on the NPS, people start asking for points. “If you were happy, please give us a 10, anything lower counts as a fail for us” is a sentence that destroys the metric. Reichheld named the problem himself in the Harvard Business Review in 2021 under the title “Net Promoter 3.0” and proposed a counterweight from the accounts, the Earned Growth Rate. That measure tracks how much revenue growth genuinely comes from returning customers and the people they refer.

Only part of the customer base gets asked

Sending the survey to your newsletter list means asking the people who already pay attention. Writing only to active accounts means asking the people who have not left yet. Both routes lift the score without anyone rigging a thing. Whether your result holds for your whole customer base is decided by who ends up in the survey, and the article on representativeness in online surveys sets out what that takes.

Improving the NPS: from a number to an action

Improving the NPS means closing a loop rather than chasing the figure: you ask, you read what people wrote, you change something that caused the complaints and you tell the people who complained what changed. Skip the last step and the next wave comes back with the same answers plus a little more resignation.

Three steps do most of the work. Read the open answers of your detractors first and sort them into causes rather than topics, because “delivery” is a topic and “the delivery date moves after the order is confirmed” is a cause. Pick one cause per quarter, the one that shows up most often, and fix that single thing properly instead of launching five small initiatives. Write back to the people who raised it and say what has changed, because a customer who gets an answer moves groups more often than one who gets a discount.

The 7s and 8s are where the quickest gains sit and they are the group almost everybody ignores. Passives leave no mark on the score, so a company with a large middle can work hard on its detractors and watch the NPS barely move. Ask the passives the same open question and the answers come back specific and small: one step too many at checkout, a reply that took two days, a feature that nearly does what they need. Moving a passive up to a 9 lifts the score by exactly as much as lifting a detractor into the passive band and it is usually the easier of the two.

Conclusion

The Net Promoter Score is one of the few metrics that everyone in the team grasps immediately, and that is its greatest strength. Treat it as an early warning system and a way into a conversation, never as proof. What matters is not the score itself but what sits in the open answers underneath it, and whether the figure moves across several waves measured the same way. A lone NPS with no shares, no sample size and no point of comparison says almost nothing.

Where to go next

• Planning the survey that goes with it? Customer surveys: guide and examples
• Answers already in? Evaluating a questionnaire in five steps
• Wondering why an average is the wrong tool here? Ordinal scale explained simply

Want to measure the NPS in a survey of your own?

empirio.ai has NPS as its own question type and the analysis works the score out for you. See the customer feedback templates