Net Promoter Score (NPS): How It Is Calculated and Misread
 - Arfadia - r6-20261015.jpg "Net Promoter Score (NPS): How It Is Calculated and Misread - Arfadia")
Net Promoter Score (NPS) is a survey score built from one question: how likely a respondent is to recommend a defined company, brand, product, or service to another person, rated from 0 to 10. Answers of 9 or 10 count as promoters, 7 or 8 as passives, and 0 to 6 as detractors. NPS is the percentage of promoters minus the percentage of detractors, so it always falls between -100 and +100.
That simplicity is the appeal. It also explains most of the mistakes. NPS records what people say they would do, not what they did, and one number can hide very different groups of customers.
What NPS actually measures
The standard item asks respondents how likely they are to recommend the named entity, on a 0 to 10 scale. Bain & Company, the originator of the method, sets out the question and the scoring bands in its guide to measuring your Net Promoter Score. Everything else in NPS is arithmetic applied to those answers.
Intention, not behavior
A rating of 10 is a statement about likelihood. It is not a recorded recommendation, a referral, a repeat purchase, or a renewal. Someone can rate a brand 10 and never mention it, while someone who rates it 6 may still renew. Intentions are quick to collect, but any claim about what customers actually did needs behavioral data next to the survey.
The labels deserve the same caution. The 9 to 10, 7 to 8, and 0 to 6 cutoffs are conventions for summarizing a group, not a test of how any single respondent will act.
The question sets the scope
The score belongs to whatever the question names. Asking about a company, one product line, or yesterday's support call produces three different measures, even if all three are labeled NPS. The people asked matter too. Current customers and everyone who recently bought in the category are different populations, so their scores are not interchangeable.
NPS, the Net Promoter System, and naming
Net Promoter Score is the single statistic; the Net Promoter System is the broader practice of collecting feedback and acting on it. Since the S already means score, adding score again is redundant. Bain's trademarks and licensing page discourages that phrasing and sets out attribution requirements for its marks.
How to calculate NPS
The calculation has three steps.
- Keep the valid answers to the recommendation item, meaning ratings from 0 to 10. Missing or invalid answers are excluded. They are not counted as passives.
- Count promoters (P, ratings of 9 or 10), passives (A, ratings of 7 or 8), and detractors (D, ratings of 0 to 6). Their total is n.
- Subtract the detractor share from the promoter share: NPS = 100 × (P - D) / n.
That is the unweighted form. If a survey applies weights, the shares become weighted shares, and the weighting method should be disclosed with the result.
Why passives stay in the denominator
Passives never appear in the numerator, which tempts some teams to drop them. That changes the answer. Passives are valid responses, so they belong in n. The reverse error happens too: counting blank answers as passives enlarges the denominator with people who never answered, pulling the score toward zero.
A worked example (hypothetical)
Imagine 200 valid ratings: 90 promoters, 60 passives, and 50 detractors. Promoters make up 45 percent and detractors 25 percent, so NPS is 45 minus 25, or +20. The counts are invented for illustration.
How to read the number
Points, percentages, and precision
The component shares are percentages. The score, and any change in it, is expressed in points. A move from +20 to +25 is a 5-point increase, not a 5 percent increase. NPS is also not an average of the original ratings. A survey where everyone answers 8 has a mean of 8 and an NPS of 0, while a survey split evenly between 9s and 7s has the same mean and an NPS of +50.
With a fixed, unweighted sample, the score can only take certain values. Each respondent who moves one category is worth 100/n points, so with 200 responses the score moves in steps of 0.5 points. In the example above, if one detractor had answered 7 instead, the score would rise from +20 to +20.5. A rounded dashboard can hide that change, and extra decimals do not make the estimate more precise.
What zero and negative values mean
Zero means the promoter and detractor shares are equal. It does not mean everyone was neutral, because a hypothetical survey with 40 percent promoters and 40 percent detractors also scores zero. A negative score means detractors outnumber promoters. Reaching -100 or +100 requires every respondent to be a detractor or a promoter, respectively.
Same score, different customers
NPS sets the passive share aside and collapses the rest into one difference, so very different response patterns can produce the same result. Brendan Rocks' paper on interval estimation for the Net Promoter Score makes the point that equal scores can rest on different distributions with different uncertainty. The hypothetical table below shows two surveys of 200 people that both score +20.
| Survey (hypothetical) | Promoters (9 to 10) | Passives (7 to 8) | Detractors (0 to 6) | NPS |
|---|---|---|---|---|
| Survey A | 90 (45%) | 60 (30%) | 50 (25%) | +20 |
| Survey B | 60 (30%) | 120 (60%) | 20 (10%) | +20 |
Survey A has many high ratings and many low ones. Survey B has fewer promoters, far fewer detractors, and most people in the middle. Those patterns suggest different questions to investigate, and the single number points to neither. That is why the component shares belong next to the score in every report.
Benchmarks need matched designs
The evidence reviewed for this entry does not support a universal, industry-independent threshold for a good NPS. A benchmark helps only when it uses the same question, the same kind of respondents, the same survey form, and a similar period; otherwise a gap may reflect design rather than customers. Your own score over time, under an unchanged design, is usually the fairest comparison.
Relationship, experience, and benchmark surveys
When and why the question is asked matters as much as how it is scored. Bain describes three forms of Net Promoter feedback. Relationship surveys are not triggered by a specific event. Experience surveys follow a defined action, such as a purchase or a support contact. Competitive benchmark surveys are designed to compare a brand with others. Bain advises against comparing relationship and experience results directly.
Qualtrics uses the terms relational and transactional and, in its guide to NPS, advises against using NPS as a transactional, post-interaction measure. That is one organization's recommendation, not an industry-wide rule. Decide which form you run, keep it consistent, and label it.
Who answered, and how sure you can be
Sampling frame, timing, and nonresponse
An NPS describes the people who were eligible, were invited, and chose to answer, at the moment they answered. Each layer can shift the result. Bain's article on creating a reliable metric discusses how sampling choices, survey mode, and the self-selection of respondents can bias scores. A report should state who was eligible, how they were invited, the channel, the field dates, and how many responded. A high response rate is welcome, but it does not prove that respondents resemble the people who stayed silent. If the channel or eligible group changes between waves, flag the break instead of reading it as a trend.
Sampling uncertainty is a separate problem
Even a perfectly drawn sample carries random error. NPS is a difference between two shares taken from the same respondents, so its uncertainty depends on the full mix of promoters, passives, and detractors as well as on sample size. Rocks compares interval methods, including adjusted Wald and score approaches, which are studied options rather than one required procedure. An interval covers sampling error only. It does not correct a biased frame, and no universal minimum sample size suits every purpose.
What the research says about NPS and growth
The metric originates with Frederick F. Reichheld's December 2003 Harvard Business Review article, The One Number You Need to Grow. The title states the proposition plainly. It is the originator's argument, though, and later studies that tested the link reached mixed, conditional results.
The Norwegian replication
Keiningham and colleagues tested the growth claim in a 2007 Journal of Marketing study. Using a dataset covering 21 Norwegian firms, with a 17-firm industry subset, they did not replicate the clear superiority claimed for NPS within the comparisons they examined. Their survey used a 1 to 10 recommendation scale rather than the standard 0 to 10 item, which limits how directly the result transfers.
The U.S. sportswear panel
A later study by Baehre, O'Dwyer, O'Malley and Lee, published in the Journal of the Academy of Marketing Science (online 2021, in print 2022), took a different angle. It used brand-health survey data on seven U.S. sportswear brands from 2013 to 2017, collected from eligible recent sportswear buyers aged 16 to 30 rather than only each brand's customers. Changes in NPS showed conditional, near-term value for predicting sales growth, while static NPS levels did not. In the same models, brand consideration predicted about as well as NPS.
Two limits matter. A predictive design does not show that raising NPS causes revenue to grow, and a comparable rival measure means NPS is not established as uniquely the best predictor. The two studies ask different questions of different samples, so they do not cancel out. NPS can carry some forward-looking signal in some settings. It is not a guaranteed growth lever.
A portfolio, not a single number
In service-sector questionnaire research, Pollack and Alexandrov examined the nomological validity of the recommendation question and recommended using NPS within a portfolio of customer feedback measures rather than as a standalone diagnostic. A related analysis by part of the sportswear team, comparing NPS with alternative calculation methods, reports that top-box measures outperformed conventional NPS within its analysis. It shares authors and possibly data with the sportswear study, so it is not an independent replication.
Using follow-up reasons without overreading them
Many NPS surveys add an open question asking why the respondent chose that rating, a practice Bain's reliability article treats as part of the method. The comments are useful. They surface customer language, unexpected problems, and themes worth testing. The trap is assuming the most frequent complaints are the changes that will raise the score.
Lance A. Bettencourt and Mark B. Houston examined that assumption in The Untested Assumption, published in the International Journal of Market Research. Across two vehicle-related research settings, the abstract reports only low to moderate agreement between priorities drawn from open-ended reasons and priorities derived or stated through other methods. Treat comments as hypotheses. Check them against behavioral data or a structured follow-up before committing budget to a fix.
NPS compared with nearby concepts
Clear boundaries keep a survey intention from being reported as something it is not.
- Customer retention is observed continued activity. A recommendation rating is not a retention rate.
- Brand loyalty is a broader attitudinal and behavioral construct. NPS can be one indicator in a loyalty measurement plan, not the definitive loyalty measure.
- Customer experience (CX) is the management domain covering every touchpoint. NPS is one survey input to that work.
- Word of mouth is actual conversation between people, favorable or not. A promoter rating does not show that any conversation happened.
- Referral marketing is an organized program, with or without rewards. NPS does not count enrollments, referred leads, or attributed conversions.
- Social proof is influence from other people's visible choices or endorsements. A private survey rating is not a public endorsement.
- CSAT and CES ask about satisfaction and effort. They answer different questions, and none is automatically the better metric.
- Voice of the customer (VoC) is the wider program of surveys, comments, and behavioral data.
Employee NPS, which asks staff about recommending an employer, is a separate variant outside this entry.
Private detractor ratings are not public reviews, but the themes behind them can inform online reputation management. For measurement that combines survey results with channel performance, see Arfadia's internet marketing services.
How to report NPS responsibly
- State the exact question wording and what it named.
- Give n, the field dates, the channel, and who was eligible.
- Show promoter, passive, and detractor shares beside the score.
- Say whether it is a relationship, experience, or benchmark measure.
- Disclose any weighting, and compare only like with like.
- Pair the score with behavioral measures before drawing business conclusions.
Frequently Asked Questions
What is Net Promoter Score (NPS)?
Net Promoter Score (NPS) is a survey-derived score of stated likelihood to recommend a defined company, brand, product, or service. It equals the percentage of promoters (ratings of 9 to 10) minus the percentage of detractors (0 to 6), calculated over all valid 0 to 10 answers, passives included. It reflects stated intention, not recorded referrals, retention, or satisfaction.
How do you calculate NPS?
Divide promoters (9 to 10) and detractors (0 to 6) by all valid responses, passives included, then subtract the detractor percentage from the promoter percentage. With 90 promoters, 60 passives, and 50 detractors out of 200, the hypothetical result is 45 minus 25, or +20.
What does an NPS of zero or below zero mean?
Zero means promoters and detractors make up equal shares of valid responses. It does not mean everyone was neutral. A negative score means detractors outnumber promoters. Neither is a verdict on business quality without context.
Is NPS a percentage?
No. The promoter and detractor shares are percentages, but NPS is their difference, expressed in points on a scale from -100 to +100. A change from +20 to +25 is a 5-point increase.
What counts as a good NPS?
The evidence reviewed here does not support a universal, industry-independent threshold. Compare only results with the same question, respondent frame, survey form, and period, such as your own score over time.
How is NPS different from CSAT and customer retention?
NPS asks about likelihood to recommend. CSAT asks about satisfaction, and customer retention is observed behavior, such as customers staying or renewing. A high NPS does not prove either one.
Does a higher NPS cause revenue growth?
Published research does not establish that. A Norwegian study did not replicate the claimed superiority of NPS, and a U.S. sportswear study found that changes in NPS had conditional, near-term predictive value without showing cause.
