What is a good NPS score? Above 0, and then it gets complicated

A good NPS score is anything above 0, and Survicate calls 30+ doing well. Why published benchmarks disagree and how to check your own score before comparing.

Instant Expert EditorialPublished Updated 5 min read

A good NPS score is anything above 0, because it means you have more promoters than detractors. Past that, "good" depends on your industry, whether you sell to businesses or consumers, the countries you survey and how you asked. For one reference point, Survicate's 2025 benchmark found a median NPS of 42 overall, 38 for B2B companies and 49 for B2C companies. Survicate calls 30 or more "doing well" and above 70 "among the best."

The catch: published benchmarks disagree, and a score from a small sample can move 20 points by chance. Your most useful benchmark is your own score over time, measured the same way, counted only when the change beats your margin of error. (New to the metric? Start with Net Promoter Score.)

Two benchmarks, two different answers

Two survey-software vendors publish benchmarks from their own customers' surveys:

SourceWhat it measuredSelected figures
Survicate, 2025599 companies, 2,187 surveys, about 5.4 million responses from January to October 2025. Surveys with 100+ responses only. Medians.All industries 42; B2B 38; B2C 49; Manufacturing 65; Healthcare 61; Agency and Consulting 59; Professional Services 50; Fintech 46; Software 30 (B2B software 29, B2C software 47)
Retently, 2026Its own customers; industries with more than 10 clients; at least 10,000 surveys; not segmented by company size or country. Averages.Financial Services 68; Consulting 68; B2B Software and SaaS 41; Internet Software and Services 26

Survicate's bottom 10% of companies scored -4 in 2025 and its top 10% scored 78. An industry median is the middle of a very wide road. Bain, which developed NPS, offers certified competitive benchmarks through services including NPS Prism.

Why B2B software scores 29 in one report and 41 in the other

Similar labels, 12 points apart. Each vendor reports on companies that chose its product, not a random sample. Survicate reports medians, which it says are less affected by extreme scores, and Retently reports averages. The two define "software" differently, and the surveys inside each dataset vary in timing, channel and audience.

Country matters too. NN/g summarizes a 2021 Qualtrics XM Institute study of 17,509 consumers across 18 countries: rating companies they liked, respondents in India and Mexico produced scores of 60 or more, and respondents in Japan produced -47. A company with mostly Japanese customers could be doing well with a score that would look alarming elsewhere.

So does the type of score. Bain separates relationship, experience and competitive benchmark scores and says relationship and experience scores should not be compared directly. A survey sent right after a resolved support ticket is not comparable with an annual survey of all customers.

Check your own score before comparing it

1. Check the math. On the 0 to 10 scale, 9 and 10 are promoters and 0 to 6 are detractors. Keep passives in the total. Dropping them inflates the score, which is exactly why it is tempting.

2. Work out the margin of error. A made-up example: a B2B software company gets 40 responses, with 18 promoters, 14 passives and 8 detractors. Its NPS is 18 ÷ 40 = 45% minus 8 ÷ 40 = 20%, or 25. Using the adjusted-Wald method that MeasuringU recommends:

  • Adjusted sample: 40 + 3 = 43
  • Adjusted promoters: 18.75 ÷ 43 = 0.436
  • Adjusted detractors: 8.75 ÷ 43 = 0.203
  • Adjusted NPS: 0.436 − 0.203 = 0.233
  • Variance: 0.436 + 0.203 − 0.233² = 0.585
  • Standard error: √(0.585 ÷ 43) = 0.117
  • 95% margin of error: 1.96 × 0.117 = 0.229, or about ±23 points

The true score could plausibly be anywhere from about 0 to 46, which covers Survicate's 29 and 38 and Retently's 41. With 40 responses, "above industry average" stays off the slide.

3. Check who answered. Bain notes that unhappy customers may not bother to respond. A score built mostly from your happiest accounts will look better than your customer base as a whole, so compare response rates across segments.

4. Write down how you asked. A survey after a good onboarding call scores differently from one at renewal. Record the trigger, channel and audience next to the number.

5. Look for gaming. NN/g's examples include staff mentioning the survey only after good interactions and coupons offered for a rating. If a bonus depends on the score, look closely.

6. Keep the method fixed. AAPOR puts it well: "If you want to measure change, don't change the measure."

The reasons matter more than the number

Survicate itself recommends comparing against your own previous results after the first survey. Then group the answers to "What is the main reason for your score?" A 25 driven by slow support calls needs different work from a 25 driven by missing features. Call a few detractors, passives and promoters to hear the rest; customer survey questions has a short follow-up call guide.

Your next step

Before you share your score, write three things next to it: the number of responses, the margin of error, and the benchmark you used (who published it and how it was measured).

A benchmark can't tell you how customers of competing products see the category. Conversations can. On Instant Expert you describe the people, such as "operations leads who switched help desk software in the last year," and review who it finds before any invitation goes out. You set what each person receives and our fee (20% of what you pay) is added on top: a $100 offer for a 30-minute call costs you $125, charged only if they book. For one question, such as "What made you leave your last help desk tool?", ask for a written or voice answer instead. They have 7 days to respond, and declines cost nothing. The directory page for contact center operations experts shows the kind of people you can reach.