---
title: "What is a good NPS score, and how do you calculate NPS?"
description: "To calculate NPS, subtract the share of detractors (0 to 6) from the share of promoters (9 and 10). Then compare the score with your industry average for 2026."
url: "https://raconte.ai/en/blog/net-promoter-score"
---

# What is a good NPS score? How to calculate it, with 2026 benchmarks

To calculate NPS, subtract the share of detractors (0 to 6) from the share of promoters (9 and 10). Then compare the score with your industry average for 2026.

October 28, 2026

Godefroy

Updated on November 19, 2026

Key takeaways

*   A good NPS score is above 0 (more promoters than detractors), solid above 30, excellent above 70, always judged against the industry average: 2026 averages run from 26 (internet software and services) to 68 (financial services).
*   NPS subtracts the share of detractors (scores 0 to 6) from the share of promoters (9 and 10). On 200 responses, 45% promoters and 25% detractors give a score of 20.
*   Every benchmark measures its own customer base: Retently, Survicate and Qualtrics disagree by 10 points or more on the same industry, so pick one source and stay with it.
*   NPS tracks a trend well. It never explains the score, and it predicts growth no better than classic satisfaction measures.
*   Call every detractor within 24 hours: a five-minute interview returns the reason for the score, which the free-text box almost never gives.

A good NPS score sits above 0, which means your promoters outnumber your detractors. It becomes solid above 30 and excellent above 70, and the only way to judge it fairly is against your industry: the 2026 averages run from 26 in internet software and services to 68 in financial services, so the same score can lead one market and trail another. This page shows how to calculate NPS step by step with a worked example, the benchmark table by sector, the limits every NPS number carries, and the follow-up that turns a score into a reason.

We build [Raconte](/en), a tool for AI-driven voice interviews, so the last part argues a simple point: the score measures, the conversation explains.

## The NPS question and its three categories

Fred Reichheld, a consultant at Bain & Company, published the method in the Harvard Business Review in December 2003, in an article titled “The One Number You Need to Grow”. He was looking for the survey question that would best predict a company’s growth. After two years of matching survey answers to what customers actually did, purchases and referrals included, the recommendation question predicted growth better than the satisfaction and return-intent questions. The idea behind NPS comes from there: a customer who takes the risk of recommending you to a friend tells you more about their loyalty than one who says they are satisfied. The acronym is still a registered trademark, held by Bain and Satmetrix, which has not slowed adoption: by 2020, two thirds of the Fortune 1000 used some version of it.

The question asks “How likely are you to recommend \[company\] to a friend or colleague?”, on a scale from 0 (not at all likely) to 10 (extremely likely), and each answer sorts the respondent into a category:

Category

Scores

What to expect

Promoters

9 and 10

They come back and talk about you

Passives

7 and 8

Satisfied but unenthusiastic, they leave for the next better offer

Detractors

0 to 6

Unhappy, and they talk about you too, badly

The same formula serves employees under the name eNPS, with the question “How likely are you to recommend this company as a place to work?”. You meet it in [employee surveys](/en/blog/employee-survey-questions). On the customer side, NPS measures intent to recommend, CSAT measures satisfaction after one specific interaction, and CES measures the effort a journey demanded: [the three metrics compared](/en/blog/customer-satisfaction-survey) help you pick the one that answers your question.

## How to calculate NPS score, step by step

The NPS formula fits on one line: NPS = percentage of promoters − percentage of detractors. Passives count in the total of responses, which is the denominator of both percentages, but they never enter the subtraction.

The calculation takes four steps:

1.  Ask the recommendation question on a 0 to 10 scale, and count every response you received.
2.  Sort each response: 9 and 10 are promoters, 7 and 8 are passives, 0 to 6 are detractors.
3.  Divide the number of promoters, then the number of detractors, by the total of responses, passives included.
4.  Subtract the percentage of detractors from the percentage of promoters. The score is written without a % sign, between −100 and +100.

Take 200 responses to a quarterly survey: 90 scores of 9 or 10, 60 scores of 7 or 8, 50 scores of 6 or below. Promoters are 90 ÷ 200 = 45% of responses, detractors 50 ÷ 200 = 25%.

Scores

Category

Responses

Share

9 and 10

Promoters

90

45%

7 and 8

Passives

60

30%

0 to 6

Detractors

50

25%

NPS = 45% − 25% = **20**

Two mistakes show up in dashboards. The average of the scores is not the NPS: an average of 8 can mean an NPS of 0, if everyone scored 8, or of 60, if 80% of respondents scored 10 and the rest 0. Drop passives from the denominator and the score inflates: in our example it goes from 20 to 29 with the same responses.

Sample size matters too. Each customer crossing a category boundary moves the score by roughly 100 divided by the number of responses: a full point on 100 responses, half a point on 200. On 20 responses, a single customer sliding from 7 to 6 takes 5 points off. At that size, one angry account sinks the score as much as a real problem would.

Two more cases often trip people up:

*   With no detractors at all, NPS equals the share of promoters. On 40 responses with 30 promoters, 10 passives and no detractors, it comes to 75. It only reaches +100 when every respondent gives a 9 or a 10.
*   Round the final score, never the intermediate percentages. With 7 promoters and 2 detractors out of 15 responses, 46.7% − 13.3% gives 33. Round each share first and you get 47 − 13 = 34, a point gained without a single extra response.

To get the score in a spreadsheet, put one score per row in column A, from A2 to A201 for the 200 responses of the example, then enter the formula `=(COUNTIF(A2:A201,">=9")-COUNTIF(A2:A201,"<=6"))/COUNT(A2:A201)*100`.

![Calculating NPS from 200 responses: 45% promoters minus 25% detractors gives a score of 20](/.netlify/images?url=_astro%2Fcalcul-nps-en.BHeFYpKa.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

## What is a good NPS score, by industry

Above 0, your promoters outnumber your detractors: the score is good. Above 30, it is solid. Above 70, your customers recommend you unprompted. Those markers only mean something next to an industry comparison, because averages differ by a factor of almost three from one sector to the next.

Industry

Average NPS 2026

Financial services

68

Consulting

68

Technology and services

63

Ecommerce and retail

61

Digital marketing agencies

49

Property management

47

Insurance

46

Logistics and transportation

42

Construction

42

B2B software and SaaS

41

Communication and media

39

Healthcare

37

Cloud and hosting

30

Internet software and services

26

Retently, an NPS survey vendor, published these averages in April 2026 from its own customer base, for industries with more than 10 clients and at least 10,000 surveys ([source](https://www.retently.com/blog/good-net-promoter-score/)). Treat them for what they are: averages from a single vendor, skewed B2B, not a norm. Survicate publishes medians from 599 companies and 5.4 million responses, and puts software at 30 where Retently shows 41 for B2B SaaS. Qualtrics, surveying more than 10,000 US consumers, placed big box retail at 37 and utilities at 13 in the third quarter of 2025. Pick one source, keep it quarter after quarter, and compare yourself to yourself.

Geography counts as well. European respondents score more conservatively than American ones, and an 8 often stands in for a glowing review. A company in France judging itself against worldwide averages grades itself on a harsher scale than its customers use.

Retently’s summary rule holds: the only good score is the one that tops your previous one. A steady climb, measured with exactly the same question each time, tells you more than the distance to an industry average.

## A number never says why

The structural limit of NPS fits in one sentence: three customers can give the same score for opposite reasons, and the formula adds them without telling them apart. An 8 because the product is excellent but the price is too high, an 8 out of habit with no formed opinion, an 8 after an incident the support team handled well: the same score calls for three different decisions, and the number shows none of them.

The subtraction only keeps the boundaries. A customer sliding from 10 to 9 changes nothing in the score, a customer sliding from 7 to 6 drops it. You will know the score moved, never whether an excellent customer got disappointed or an indifferent one turned unhappy. Those two movements call for different responses.

The metric was contested from publication. Timothy Keiningham, a researcher at Ipsos Loyalty, and three co-authors replayed the original demonstration on 21 firms and more than 15,500 interviews from the Norwegian Customer Satisfaction Barometer. Their study, published in the Journal of Marketing in 2007, found that NPS predicts revenue growth no better than classic satisfaction measures, including in the industries Reichheld had cited as exemplars. NPS keeps its place as a tracking metric. A strategic decision needs something else: a method that gives the reason, picked by where the customer stands, from the [customer research methods](/en/blog/customer-research-methods) that run from purchase to cancellation. [The other blind spots of a survey score](/en/blog/customer-satisfaction-survey) have their own article.

## The open question respondents skip

The open question after the score is the only one that could explain it, and it is the least filled in the survey. The Pew Research Center measured, across 92 open-ended questions in more than 40 surveys of its American panel, [18% nonresponse against 1% to 2% on closed questions](https://www.pewresearch.org/decoded/2021/10/14/why-do-some-open-ended-survey-questions-result-in-higher-item-nonresponse-rates-than-others/). Pew’s panel is paid to answer: on a customer survey sent cold, the gap widens.

The answers that do arrive run three words long. “Too expensive”, “nothing to report”, “slow support”: none of them says too expensive compared to what, or since when, or what tipped the opinion. A form cannot follow up on its own respondent, and the follow-up is exactly what turns a score into an explanation a product or support team can act on.

## Follow up with detractors within 24 hours

A detractor just told you, in one number, that they have a story. Get back to them fast: Bain, the firm behind the Net Promoter System, recommends contacting unhappy customers within 24 hours, while the experience is fresh. Companies like Charles Schwab and Apple Retail call their detractors the same day, with the response context in front of them.

Bain splits that follow-up into an inner loop, the reply to the customer, and an outer loop, the fix for everyone: a [voice of the customer program](/en/blog/voice-of-the-customer) covers both.

Five questions turn a score into a reason, anchored in what happened rather than in what would happen:

*   “Walk me through your last use of \[product\]. What happened, step by step?”
*   “What produced your score of \[N\]?”
*   Pick up the vague word from their answer as a question: “too expensive” becomes “too expensive compared to what?”
*   “What would have changed your score?”
*   “What did you use before, and what brought you here?”

Rob Fitzpatrick, author of _The Mom Test_, gives the rule: ask people about what they did, never about what they would do. The wordings that dig into a vague answer are grouped in our [open-ended questions for qualitative research](/en/blog/open-ended-questions-qualitative-research).

Five minutes per detractor is enough, but multiplied by every detractor of the wave, it becomes a slot to book, an interviewer to free up and notes to write up. Most teams plan this follow-up, then drop it halfway through the quarter.

To choose the form of that follow-up, [compare the survey, the human-run interview and the asynchronous voice interview](/en/blog/questionnaire-ou-interview-vocale): it all fits in one table.

We built [Raconte](/en) for that follow-up. You write the interview once as a prompt, send a link to each detractor, and the person talks for a few minutes whenever it suits them, in their browser. The AI listens, follows up as soon as an answer stays vague, and stops once it has asked everything. You get the transcript, the audio, a summary and the sentiment of each message within seconds of the conversation. The [pricing](/en/pricing) is simple: a five-minute interview costs €0.50 beyond the 60 free minutes of every month, no card required.

```mermaid
graph TD
  A["The NPS survey returns each quarter"] --> B["Every detractor receives an interview link"]
  B --> C["Five minutes of conversation return the reason for the score"]
  C --> D["The transcript goes to the team concerned"]
  D --> E["The customer gets a reply within 24 to 48 hours"]
  E --> F["The fix enters the monthly review"]
```

## Where to start

Keep the survey for the trend: the same question, to the whole base, every quarter, with the wording unchanged. Handle detractors as they arrive: each month, an interview link goes to everyone who scored 0 to 6 since the last wave, and you read the transcripts before the product review. Reply to every detractor, even briefly, because someone who took the time to explain deserves to know it was read.

NPS measures, the interview explains. Together they form the smallest customer tracking setup that works: [create a free Raconte account](/en) for the second half, the first 60 minutes of every month are free.

## Frequently asked questions

### What is a good NPS score?

Above 0 is good: your promoters outnumber your detractors. Above 30 is solid, above 70 is excellent. Judge the score against its industry: 2026 averages run from 26 in internet software and services to 68 in financial services, so a score of 30 can lead one market and trail another. A good score is one that tops your previous measurement.

### How do you calculate NPS score?

NPS equals the percentage of promoters minus the percentage of detractors. Count the promoters (scores 9 and 10) and the detractors (scores 0 to 6), turn each group into a percentage of all responses, then subtract. On 200 responses, 90 promoters make 45% and 50 detractors make 25%, so NPS = 45 − 25 = 20. Passives (scores 7 and 8) only count in the total of responses.

### What does NPS stand for?

NPS stands for Net Promoter Score: the net share of customers willing to recommend the company, with detractors subtracted. The acronym covers the question, the score it produces and the follow-up method around it. It is a registered trademark held by Bain & Company and Satmetrix.

### What is the difference between NPS and CSAT?

NPS measures intent to recommend the company, a relationship tracked over time. CSAT measures satisfaction after one specific interaction, a purchase or a support ticket, on a 1 to 5 scale. NPS goes to the whole base every quarter, CSAT goes out hot after each event. CES, the third metric, measures the effort a journey demanded.

### Should NPS be 0 to 10 or 1 to 10?

Use the 0 to 10 scale, which the original method and every published benchmark rely on. The bands are promoters at 9 and 10, passives at 7 and 8, detractors at 0 to 6, and a 1 to 10 scale shifts those boundaries. Changing the scale also breaks comparability with your own history and with industry benchmarks, so keep 0 to 10 and never change the wording.

### What is eNPS?

eNPS (employee Net Promoter Score) applies the same formula to employees, with the question “How likely are you to recommend this company as a place to work?”, scored 0 to 10. Promoters are the employees who would recommend the company as an employer. You find it in engagement surveys and internal barometers.

Ready to launch your first interview?

Describe your interview, share the link: an AI agent runs it by voice with each participant. You get back the transcript and analysis.

[Try for free](/login)

## Continue reading

[![Customer satisfaction survey: an example and its limits](/.netlify/images?url=_astro%2Fthumbnail.BVaPkpjh.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

August 19, 2026

## Customer satisfaction survey: an example and its limits

A customer satisfaction survey template you can copy, what every question tells you, and the four blind spots of a score out of 10.



](/en/blog/customer-satisfaction-survey)[![Voice of the customer (VoC): definition, examples, template](/.netlify/images?url=_astro%2Fthumbnail.BE9tw1Sw.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

October 13, 2026

## Voice of the customer (VoC): definition, examples, template

What VoC means, examples from tickets, reviews, surveys and interviews, and a lightweight program template to copy, with no enterprise platform.



](/en/blog/voice-of-the-customer)[![Product/market fit survey: questions, template and scoring](/.netlify/images?url=_astro%2Fthumbnail.7H1ssbYs.jpg&w=1200&h=630&dpl=6b031bf5959953076155d961)

October 22, 2026

## Product/market fit survey: questions, template and scoring

The Sean Ellis question, who to send it to, Superhuman's four questions, how to read the 40% threshold, and a survey template to copy.



](/en/blog/product-market-fit-survey)
