What Is NPS? Net Promoter Score Formula, Calculation and Benchmarks

Most people look up NPS because someone handed them a number like +32 and they need to know whether it is good, or because they are about to send their first survey.

This guide gives the definition and origin, the formula with three worked calculations, benchmarks with the method behind each, how to run the survey and how many answers you need, how NPS differs from CSAT and CES, and its criticisms. Every source was checked on 12 November 2026.

What is NPS (Net Promoter Score)?

NPS, short for Net Promoter Score, is a single number that compares how many of your customers would recommend you with how many would not. Customers answer one question on a 0 to 10 scale, and the score is the share who answer 9 or 10 minus the share who answer 0 to 6.

The word "net" is the point: critics are subtracted from fans, so mildly content customers cannot make a company look loyal. Bain & Company, which created the score, calls it "your customer balance sheet" on its Net Promoter System site, trackable for a whole company or for each product, store or team.

The Net Promoter System is Bain's name for the routine around the number: asking customers why they gave their score, following up with individuals and fixing the causes they name.

What does NPS stand for, and where did the Net Promoter Score come from?

NPS stands for Net Promoter Score, a metric Fred Reichheld of Bain & Company introduced in "The One Number You Need to Grow", Harvard Business Review, December 2003. It argued that complex satisfaction surveys measured the wrong thing.

Bain's history of Net Promoter says Reichheld and a Bain team looked for the best one-question indicator of customer lifetime value, testing questions against later repurchases and referrals. One question worked best for most mature, competitive industries: "How likely are you to recommend Company X to a friend or colleague?" In 11 of the 14 industry case studies, no other question predicted behaviour as well. The page says two-thirds of the Fortune 1000 now calculate the score, and that a second question, "Why?", turned it into the Net Promoter System.

Bain credits the research data to NICE Systems. Its trademarks page, checked on 12 November 2026, says Net Promoter, NPS, NPS Prism, Net Promoter System and the NPS-related emoticons are registered trademarks of Bain & Company, Inc., NICE Systems, Inc., and Fred Reichheld, and that Net Promoter Score is a service mark of the same three; some notices, such as Qualtrics XM Institute's, still name Satmetrix in NICE's place. Commercial use of the marks needs Bain's written consent through an NPS License; this is not legal advice. PepoChat is not affiliated with, licensed by or endorsed by Bain & Company, NICE Systems or Fred Reichheld; this page only explains the metric.

How do you calculate an NPS score?

You calculate an NPS score by subtracting the percentage of detractors (answers of 0 to 6) from the percentage of promoters (answers of 9 or 10). Passives (7 or 8) never add or subtract points, but they count in the total, so they dilute both percentages.

NPS = (promoters / all responses x 100) - (detractors / all responses x 100)

Promoters  = answers of 9 or 10
Passives   = answers of 7 or 8 (counted in "all responses" only)
Detractors = answers of 0 to 6

1. Count every valid answer, leaving out blanks, test accounts and staff.
2. Divide the promoters by the total to get the promoter percentage.
3. Divide the detractors by the total to get the detractor percentage.
4. Subtract the detractor percentage from the promoter percentage.
5. Report the result with its sign, such as +40 or -12, and the number of responses.

If every respondent is a promoter the score is +100; if every one is a detractor it is -100. A score of 0 means promoters and detractors are equal in number, not that nobody has an opinion.

Worked example 1: a simple calculation

An online bakery asks the question a week after delivery and gets 80 answers: 44 scores of 9 or 10, 24 scores of 7 or 8, and 12 scores from 0 to 6.

1. Total responses: 44 + 24 + 12 = 80.
2. Promoters: 44 ÷ 80 = 0.55, so 55%.
3. Detractors: 12 ÷ 80 = 0.15, so 15%.
4. NPS: 55 - 15 = +40.

The 24 passives (30% of answers) appear only in the total. Bain describes passives as satisfied for now, but says they may defect if a competitor's ad catches their eye.

Worked example 2: how passives change the outcome

Two branches of a shop each get 60 promoters and 40 detractors. Branch A has no passives, so 100 responses: 60% - 40% = +20. Branch B also has 100 passives, so 200 responses: 60 ÷ 200 = 30% promoters, 40 ÷ 200 = 20% detractors, and 30 - 20 = +10.

Same fans, same critics, half the score, because passives sit in the denominator. They also decide where Branch B goes next: if all 100 moved up to a 9, it would have 160 promoters out of 200 (80%) against 20% detractors, +60; if all 100 slipped to a 6, 30% promoters against 140 detractors out of 200 (70%), -40. The passives alone span 100 points.

Worked example 3: a negative NPS

After a month of delayed shipments, a store gets 120 answers: 18 promoters, 42 passives and 60 detractors.

1. Promoters: 18 ÷ 120 = 15%.
2. Passives: 42 ÷ 120 = 35% (used only in the total).
3. Detractors: 60 ÷ 120 = 50%.
4. NPS: 15 - 50 = -35.

Half the respondents scored 7 or more, yet the score is deeply negative: the metric treats a 6 as a warning, not a pass.

What is a good NPS score?

A good NPS is a score above 0 that is rising over time and beats the average for your industry when both are measured the same way. Published industry figures for 2024 and 2025 run from about +16 to +65, and the survey method moves the number as much as the industry does.

Three sources publish industry figures with a stated method. Qualtrics XM Institute surveyed 10,000 US consumers in the second half of 2024 about 354 companies in 22 industries and averaged the company scores in each industry (published September 2025). NPS Prism, a benchmarking service backed by Bain, reported US figures for Q1 to Q4 2025 from a double-blind survey of more than 800,000 customers covering 1,000 companies. Survicate, a survey tool, published the median scores of its customers' own NPS surveys: 599 companies and about 5.4 million responses from January to October 2025, counting only surveys with at least 100 responses a year.

US banks score +28.0 in one source and +33 in another. XM Institute and NPS Prism ask consumers about brands they deal with; Survicate's figures come from companies surveying their own customers, which tends to read higher. Bain's page on the three types of NPS explains why: people who do not like doing business with you may decide your survey is not worth their time.

Academic work shows the same gap. A 2022 study in the Journal of the Academy of Marketing Science by Baehre, O'Dwyer, O'Malley and Lee tracked seven US sportswear brands over five years: NPS among current customers averaged +43.26, and among all potential customers -3.22. Same brands, same question, a gap of about 46 points.

What is the difference between relational and transactional NPS?

Relational NPS asks customers about the company as a whole on a schedule, while transactional NPS asks right after a specific interaction such as a purchase or a support call. Bain calls them relationship NPS and experience NPS, adds a third type for competitor comparisons, and says the first two "should not be compared directly".

A post-chat survey mostly reflects that conversation; a relational survey reflects price, product, delivery and every past contact, so the two give different numbers in the same month.

How do you run an NPS survey?

To run an NPS survey, ask the standard question on a 0 to 10 scale, follow it with an open "why" question, send it to every eligible customer or a random sample at a planned moment, and collect enough answers for the score to mean something.

The exact question

Bain's wording is "How likely are you to recommend us to a friend or colleague?", or the same with your company's name. Show all eleven options from 0 to 10 and label the ends, for example "Not at all likely" and "Extremely likely". Bain says a scale starting at 0 stops people mixing up its ends, which happens on 1 to 10 scales.

Follow it with one open question, such as "What is the main reason for your score?" Bain calls "Why?" the critical second question: the reason is what you can act on.

When to send it

Send relational surveys on a fixed calendar, once or twice a year, so changes reflect customers rather than timing. Send transactional surveys soon after the interaction while it is fresh, and cap how often one person is asked; Bain's guidance is at most once every three months per customer.

Who to ask

Ask every eligible customer, or a random sample, never a hand-picked list, and leave out staff, test accounts and anyone surveyed recently. A list filtered to happy customers produces a flattering number that predicts nothing.

How many responses you need

You need a few hundred responses before small movements in NPS mean anything, because small samples swing. Expect few replies: Bain notes that any given Net Promoter feedback request may have "response rates in the low single digits", so 200 answers can mean asking several thousand customers.

The 95% margin of error for NPS uses the shares of promoters and detractors:

Margin of error = 1.96 * sqrt((P + D - (P - D)^2) / n) * 100

P = share of promoters (0.50 for 50%)
D = share of detractors (0.20 for 20%)
n = number of responses

With 50% promoters and 20% detractors, a measured NPS of +30, P + D - (P - D)² is 0.70 - 0.09 = 0.61. The table applies the formula at different sample sizes.

With 30 responses, one customer changing a 9 to a 6 moves the score by 6.7 points (2 ÷ 30 × 100), so a jump from +30 to +40 proves nothing yet. Halving the margin takes four times as many answers, and no number of responses fixes a biased list.

What is eNPS (employee Net Promoter Score)?

Employee Net Promoter Score (eNPS) applies the Net Promoter idea to staff: Bain's employee NPS page says most adopters ask "How likely would you recommend this company as a place to work?", usually scored the same way, promoters minus detractors. Bain says eNPS surveys should be a few questions long, that companies most commonly ask monthly, and that answers must stay confidential, so there are no follow-up calls. It also warns that employee scores "can be substantially lower than customer scores", so never compare an eNPS with a customer NPS.

NPS vs CSAT vs CES: what is the difference?

NPS measures loyalty to the whole company, CSAT measures satisfaction with one interaction, and CES measures how much effort the customer had to spend to get something done. A common pattern is CSAT or CES after contacts and NPS once or twice a year.

Customer satisfaction score (CSAT) is the share of customers who rate an interaction positively, such as 4 or 5 out of 5. Customer effort score (CES) comes from "Stop Trying to Delight Your Customers" by Matthew Dixon, Karen Freeman and Nick Toman (Harvard Business Review, July to August 2010), whose study of more than 75,000 people found it predicted loyalty better than satisfaction measures or NPS. PepoChat's thumbs-up or thumbs-down after a resolved chat is the CSAT kind of measure, not NPS. Customer sentiment analysis for support chats compares survey, rating and AI-based approaches.

How do you improve your NPS?

You improve NPS by fixing the reasons customers give for low scores, following up with detractors quickly, and making support easier and faster; asking customers for a high score improves only the number.

1. Close the loop with detractors. Contact every customer who scored 0 to 6 within a few days and fix their case where you can. Bain calls this "closing the loop".
2. Code the "why" answers. Tag each comment with one reason (price, delivery, a bug, a slow reply) and count them monthly. The top two reasons are your plan.
3. Work on the passives. They decide most of the movement (worked example 2); ask what would have made their 8 a 9.
4. Cut customer effort. Fewer repeat contacts and answers that solve the problem the first time matter more than delight, according to the 2010 HBR research; how to reduce customer support tickets lists the tactics.
5. Answer faster at every hour. Long waits show up in "why" comments; first response time benchmarks gives targets by channel. An AI support agent such as PepoChat answers routine questions from your knowledge base at any hour and hands the rest to a person, which shortens the wait without promising a score.
6. Never chase the number itself. Do not ask customers for a 9 or 10, tie bonuses to raw scores or survey only happy customers; habits like these are how NPS gets gamed.

What are the main criticisms of NPS?

The main criticisms of NPS are that independent researchers could not reproduce its claim to be the best predictor of growth, that the score depends heavily on who is asked, and that a single self-reported number is easy to game.

It does not reliably beat other measures

Keiningham, Cooil, Andreassen and Aksoy (Journal of Marketing, 2007) used data from 21 firms and more than 15,500 interviews in the Norwegian Customer Satisfaction Barometer, and in the industries Reichheld had cited as examples failed to replicate the "clear superiority" of Net Promoter over other measures.

The answer depends on who you ask

In the 2022 Baehre study, only NPS measured across all potential customers, buyers and non-buyers alike, predicted future sales growth; NPS measured only among current customers did not. The authors concluded that the original premise is reasonable but academics' methodological concerns are valid.

Effort may matter more

The 2010 HBR research found customer effort predicted loyalty better than satisfaction or NPS, so pair NPS with CES.

It is easy to game

In "Net Promoter 3.0" (Harvard Business Review, November to December 2021), Reichheld, Darci Darnell and Maureen Burns wrote that NPS had been gamed and misused and that unaudited, self-reported scores undermined it. They proposed a companion accounting metric, the earned growth rate, based on revenue from returning customers and referrals.

The grouping throws away detail

A 0 and a 6 count the same, and a move from 6 to 7 changes the score while a move from 7 to 8 does not. Two companies can share an NPS with very different spreads of answers, so look at the full distribution too.

What are the most common NPS calculation mistakes?

The most common NPS mistakes are averaging the 0 to 10 answers, counting passives as promoters, and pooling surveys from different channels or survey types into one number.

Averaging the scores

Ten answers of 8, 8, 8, 8, 8, 6, 6, 6, 6 and 10 average 7.4 out of 10, which sounds healthy. The NPS is one promoter (10%) minus four detractors (40%): -30. Never report an average under the name NPS.

Counting passives as promoters

In the bakery example, treating the 24 passives as promoters gives 68 ÷ 80 = 85% promoters and +70 instead of +40. Only 9 and 10 count as promoters.

Mixing channels and survey types

Say a post-chat survey gets 40 answers (30 promoters, 6 passives, 4 detractors: +65) and a relationship email gets 400 (120, 160 and 120: 0). Pooled, that is 150 promoters and 124 detractors out of 440, about +6. If chat answers grow to 400 next quarter with the same mix, the pooled score jumps to +32.5 though neither group changed. Report each survey type and channel on its own line.

Where PepoChat fits

PepoChat does not send NPS surveys. It is an AI support agent for websites, and the satisfaction signal it records is smaller and more immediate: when a conversation is resolved, the chat widget asks "How was this conversation?" with a thumbs-up and a thumbs-down.

Each vote appears as a "Rated good" or "Rated poor" badge in the team inbox. The Analytics page shows "Positive ratings", the thumbs-up share of all votes with both counts, for 7, 30 or 90 days, plus a 90-day Sentiments card; test chats are left out. The analytics documentation describes the cards, and the REST API returns each conversation's rating. It is a CSAT-style measure of one chat, with limits: visitors who leave before a chat is resolved never see the prompt, and there is no comment box.

For NPS, use a survey tool. A developer can feed it from PepoChat's signed conversation-resolved webhook, which includes the visitor's email when they gave one, within your consent rules.

PepoChat's real part is the service behind the score: answers from your own knowledge base at any hour, and a human handoff to your team when the agent cannot help. PepoChat publishes no satisfaction or NPS figures. Start a free workspace to see how visitors rate your answers; the Free plan includes 100 AI replies a month.

What to do next

• Recalculate your latest NPS by hand and check it against the margin-of-error table before calling a change a trend.
• Put NPS next to the support metrics behind it: AI support chatbot metrics that matter, how vendors define a chatbot resolution and the sourced AI customer service statistics.
• Compare PepoChat's plans on the pricing page, including what Free covers and where voice and phone start.
• Create a free PepoChat workspace to answer customers from your knowledge base at any hour and collect a rating on every resolved chat.
• Related explainers: What is CSAT?, What is an SLA? and What is a help desk?.