Net Promoter Score tells you where to look, not why. Here is how to read it
Net Promoter Score is promoters minus detractors on one 0 to 10 question. How to calculate it, find its margin of error, know its limits and call people to learn why.
Net Promoter Score (NPS) tells you where to look, not why. It comes from one question: "How likely are you to recommend [company] to a friend or colleague?" People answer from 0 to 10. Those who answer 9 or 10 are promoters, 7 or 8 are passives, and 0 to 6 are detractors. NPS is the percentage of promoters minus the percentage of detractors, so it runs from -100 to +100.
It is easy to collect, which is why it is everywhere. It is also noisy with small samples and differs by country and industry. Use it to decide whom to call.
It started as one question that predicted growth
Fred Reichheld of Bain & Company introduced the idea in a December 2003 Harvard Business Review article, The One Number You Need to Grow. The summary says it drew on two years of research linking survey answers with customer purchasing and referrals. In most industries studied, the share of customers enthusiastic enough to refer a company correlated with its growth relative to competitors.
Bain's history of Net Promoter adds that the team worked with data from NICE Systems, and that in 11 of the 14 industry case studies it compiled, no other question predicted behavior as well. Bain also says it learned the score alone was not enough: asking customers "why" in their own words is what makes improvement possible. Net Promoter and NPS are registered trademarks of Bain, NICE Systems and Reichheld.
1. Calculate it with passives in the total
Here is a made-up survey with 200 responses:
- 90 people answered 9 or 10 (promoters)
- 70 people answered 7 or 8 (passives)
- 40 people answered 0 to 6 (detractors)
The arithmetic:
- Promoters: 90 ÷ 200 = 45%
- Detractors: 40 ÷ 200 = 20%
- NPS: 45 - 20 = 25
Passives count toward the total but toward neither group (NN/g). Three common mistakes:
- Leaving passives out of the total. Using 130 as the base gives about 69% - 31% = 38, which overstates the score by more than 13 points (a pleasant afternoon, then an awkward correction).
- Reporting the average rating. An average such as 7.8 out of 10 is a different measure.
- Changing the scale. Bain says the 0 to 10 scale was chosen partly because people sometimes pick "1" by mistake on a 1 to 10 scale, thinking it is the top. Change the scale and your score is no longer comparable with anyone else's.
2. Work out the margin of error before you celebrate
A sampled NPS has a wider margin of error than most people expect, because it combines two percentages. MeasuringU researchers Jim Lewis and Jeff Sauro compared methods for NPS confidence intervals and recommend an adjusted-Wald method from Brendan Rocks's 2016 paper. Applied to the example at 95% confidence:
- Add 3 to the sample: 200 + 3 = 203
- Add 0.75 to promoters and detractors: 90.75 and 40.75
- Adjusted shares: 90.75 ÷ 203 = 0.447 and 40.75 ÷ 203 = 0.201
- Adjusted NPS: 0.447 - 0.201 = 0.246
- Variance: 0.447 + 0.201 - 0.246² = 0.587
- Standard error: √(0.587 ÷ 203) = 0.054
- Margin of error: 1.96 × 0.054 = 0.105, or about ±10.5 points
So a score of 25 from 200 responses is consistent with a true score from about 14 to 35. With the same mix from 800 people, the margin shrinks to about ±5 points. From 40 people, it is about ±23. A five-point change between quarters on 200 responses is not, by itself, evidence that anything changed.
3. Know what the number can't tell you
NN/g's review of NPS lists these limits:
- It tells you how people feel, not why. Even a follow-up "why" relies on self-report, and some people won't answer it.
- It needs a reasonably large sample. Not a handful of usability sessions.
- Grouping throws away information. Moving customers from a 2 to a 5 is real progress, but they still count as detractors.
- It measures loyalty to a whole company or product. It is a poor way to judge one screen or feature.
- It is easy to game. NN/g's examples include staff mentioning the survey only after a good call, or offering a coupon for a rating.
Scores also vary by country. NN/g summarizes a 2021 Qualtrics XM Institute study of 17,509 consumers in 18 countries: asked about companies they liked, respondents in India and Mexico produced scores of 60 or more, while respondents in Japan produced -47. NN/g recommends keeping the standard calculation and noting the context instead of adjusting the formula.
NPS isn't the only loyalty measure with research behind it. Dixon, Freeman and Toman argued in Harvard Business Review that, for service interactions, a customer effort score predicted loyalty better than satisfaction or NPS.
4. Don't compare the three types of NPS
Bain distinguishes three types of NPS and says they shouldn't be compared directly:
- Relationship NPS is asked without a trigger, often once or twice a year, to measure general sentiment.
- Experience NPS follows a specific interaction, such as a support call. Bain suggests surveying any one customer no more than once every three months.
- Competitive benchmark NPS is collected double-blind by a third party across you and your competitors. Bain notes this avoids the bias of surveying only your own customers, some of whom won't bother to answer.
What counts as a good NPS covers published benchmarks and how to read them.
5. Call people from each group to learn why
Ask "What is the main reason for your score?" right after the score. Then call a few people from each group:
- Detractors: "Walk me through the last time [product] let you down. What happened next?"
- Passives: "What would have to change for you to recommend us without hesitation?" Bain describes passives as satisfied for now but open to a competitor.
- Promoters: "Have you recommended us to anyone? What did you say?" The answer is the benefit people actually repeat.
In B2B, ask several people per account. Bain's page on NPS in B2B notes that accounts include gatekeepers, business leaders and operational users with different needs. Customer survey questions has a short survey and call guide, and customer interview questions has more wording.
Your next step
Calculate your score with passives in the total, work out its margin of error, then book five to ten calls split across detractors, passives and promoters.
Your survey only hears from your own customers. To learn how buyers who chose a competitor see the category, Instant Expert can find people who match a description you write. You review who it finds, it sends your invitations, and you pay only for calls that get booked. The directory page for product professionals in enterprise SaaS is one place to start.