Back
Net Promoter Score: What It Is, How to Calculate It, and What It Misses
SmartInterview Team
The Short Answer
Net Promoter Score is one question scored on 0 to 10: "How likely are you to recommend us to a friend or colleague?" You subtract the percentage of detractors (0-6) from the percentage of promoters (9-10). Passives (7-8) are ignored. The result runs from -100 to +100.
Formula: NPS = %Promoters - %Detractors
Promoters score 9-10. Passives score 7-8. Detractors score 0-6.
NPS is a whole number, not a percentage, even though it is built from percentages.
There is no universal "good" score. It depends heavily on your industry and the country your respondents live in.
The number tells you almost nothing on its own. The open follow-up question is where the value sits.
Need to run the math on your own data? Use the NPS calculator.
The Three Groups, Side by Side
Every respondent falls into exactly one bucket. The asymmetry is deliberate: the scale is weighted so that only near-perfect scores count as advocacy.
Group | Score | What it signals | Effect on the formula | What to do next |
|---|---|---|---|---|
Promoters | 9-10 | Loyal enough to put their own reputation behind you | Adds to the score | Ask what specifically earned the 9 or 10, then protect it |
Passives | 7-8 | Satisfied but unattached, easily poached by a competitor | Counted in the base, but adds nothing | Find the one missing thing that would move them to 9 |
Detractors | 0-6 | Unhappy, and a source of negative word of mouth | Subtracts from the score | Route to a human quickly, before the churn decision is made |
NPS Against the Other Common Metrics
NPS is not a substitute for satisfaction or effort measurement. They answer different questions.
Metric | Question asked | Scale | Best used for | Main weakness |
|---|---|---|---|---|
NPS | Likelihood to recommend | 0-10, reported -100 to +100 | Tracking overall relationship health over time | Throws away most of the scale detail |
CSAT | How satisfied were you? | Usually 1-5, reported as % satisfied | Rating one specific interaction | Skews high, ceilings out quickly |
CES | How easy was it to get this done? | Usually 1-7 | Diagnosing friction in service journeys | Says nothing about affection for the brand |
Open follow-up | Why did you give that score? | Free text or voice | Explaining any of the above | Needs coding before it scales |
How to Calculate NPS, Step by Step
Take 200 completed responses:
90 people scored 9 or 10
70 people scored 7 or 8
40 people scored 0 to 6
Promoters: 90 / 200 = 45%
Detractors: 40 / 200 = 20%
NPS = 45 - 20 = +25
Two things trip people up. First, passives stay in the denominator. You divide by every respondent who answered, not just promoters plus detractors. Second, the answer is written as 25 or +25, never 25%. It is a net difference between two percentages, which is not itself a percentage.
Why the Range Is -100 to +100
The bounds fall straight out of the arithmetic. If every single respondent gives a 9 or 10, promoters are 100% and detractors are 0%, so the score is +100. If everyone gives 0 to 6, the reverse happens and you get -100. If everyone gives 7 or 8, promoters and detractors are both 0% and the score is exactly 0, despite nobody being unhappy.
That last case is worth sitting with. A score of 0 can mean a room full of mildly content customers, or it can mean half your customers love you and half want to leave. The metric cannot tell those apart. This is the single biggest reason to keep the underlying distribution in your reporting rather than shipping the headline number alone.
What Counts as a "Good" NPS
There is no defensible universal threshold, and you should be skeptical of any article that gives you one without saying where the data came from. Two factors dominate.
Industry
Categories differ in how much emotional attachment they can generate at all. People recommend a restaurant, a game or a favorite piece of software readily. Almost nobody spontaneously recommends their electricity supplier, their bank's back-office process or their insurance claims department, no matter how competently those run. Comparing a utility's NPS to a consumer app's NPS tells you about the category, not about performance.
Region and language
Response style varies across cultures. Cross-cultural survey research has repeatedly documented that respondents in some countries use the extreme ends of rating scales more readily than others, and that this varies systematically by country and language. Because NPS only counts 9s and 10s, a population that habitually avoids the top of a scale will produce a structurally lower score for identical underlying sentiment. If you run the same survey in several markets, expect the ranking between markets to be partly an artifact of scale use. Compare each market against its own history rather than against each other.
The practical rule: your only trustworthy benchmark is your own score, measured the same way, on the same population, over time. If you want an external benchmark, buy a syndicated study that publishes its methodology and sample, and check that it collected data the same way you do.
Transactional vs Relational NPS
These are two different instruments that share a question wording, and mixing them corrupts both.
Relational NPS | Transactional NPS | |
|---|---|---|
Trigger | Time based, for example quarterly or twice a year | Event based, sent after a defined interaction |
Question frame | The brand or the overall relationship | The specific experience just completed |
Sample | A representative cut of the whole customer base | Only people who triggered the event |
Used for | Board reporting, trend lines, market comparison | Operational fixes, agent coaching, process repair |
Typical failure | Too slow to catch a live problem | Biased toward customers with recent problems |
Transactional NPS is almost always higher or lower than relational NPS for the same company, because the sample is different. Support-ticket NPS in particular over-samples people who had something go wrong. Reporting it as "our NPS" is a category error. Keep two separate trend lines and label them.
The Criticisms Worth Taking Seriously
NPS was introduced by Fred Reichheld in a 2003 Harvard Business Review article, "The One Number You Need to Grow," and developed with Bain & Company and Satmetrix. The central claim was that this one question predicted growth better than the long satisfaction batteries it was meant to replace. That claim has been contested in the academic literature ever since.
The predictive claim did not replicate
The best-known challenge came from Timothy Keiningham and colleagues in the Journal of Marketing (2007), who reanalyzed the relationship between recommend-intent metrics and firm revenue growth and reported that NPS did not outperform established satisfaction measures such as the American Customer Satisfaction Index. Later work in the same vein has argued that where NPS correlates with growth, the correlation is not uniquely strong and does not establish causation. You can use NPS as a tracking metric without accepting the original "one number you need" framing.
Bucketing throws away information
Collapsing an 11-point scale into three groups discards most of the variance in your data. A customer moving from 2 to 6 is a real improvement that NPS records as nothing at all. A customer moving from 8 to 9 is a small shift that NPS records as a full swing. Statistically, the mean of the raw 0-10 scores is a more efficient use of the same responses, and it is more sensitive to change. Many teams now track both: NPS for continuity with past reporting, mean score for detecting movement early.
The same score, very different companies
Because only the net matters, wildly different distributions produce identical scores. 40% promoters and 40% detractors gives 0. So does 100% passives. So does 20% promoters, 60% passives and 20% detractors. Those are three different businesses with three different action plans. Always publish the split alongside the score.
It is easy to game and often gamed
Once NPS is tied to bonuses, staff start coaching customers on which number to pick, surveys get sent selectively to happy customers, and detractors quietly get excluded from the send list. None of that is unusual, and all of it is invisible in the headline figure. If the score is a target, treat any improvement with suspicion until you can show the sample and the send logic did not change.
Small samples make it noisy
NPS is a difference of two proportions, so its sampling error is wider than that of a single proportion at the same base size. With a handful of responses, a two-person swing can move the number by double digits. A great deal of month-to-month "movement" in NPS dashboards is nothing but noise. Set a minimum base you will report on and hold the line.
Why the Follow-Up Question Matters More Than the Score
The score tells you the direction. The follow-up tells you what to do on Monday. If you only have budget for one improvement to your NPS program, spend it here.
Ask a specific "why," not a generic one
"Why did you give that score?" pulls thin answers like "good service." Better prompts are anchored to the score the person just gave:
To detractors: "What went wrong, and what would have needed to happen for this to go well?"
To passives: "What is the one thing that would have made this a 9 or 10?"
To promoters: "What specifically do you tell other people about us?"
The promoter question is the underused one. It gives you your own positioning language, written by the people who already buy.
Probe once, then stop
Most open NPS answers stop one layer above the actual cause. "Delivery was slow" is a symptom. One follow-up question, asked while the person is still in the survey, usually gets you to the specific: which order, which promise, what they expected instead. This is where AI follow-up probing earns its place. A model that reads the first answer and asks one targeted second question turns a one-line comment into a usable account, without a moderator on the call. Voice makes this better still, because people say far more out loud than they will type into a box. We covered the mechanics of that in voice surveys vs traditional surveys.
SmartInterview handles this pattern directly: respondents answer the NPS question, then an AI follow-up probes the reason in their own language, and the open responses are coded automatically so the themes are countable instead of sitting in a spreadsheet nobody opens.
Code the answers, or the program dies
Open text only survives contact with a business review if it is quantified. You need to be able to say "shipping delays account for 31% of detractor comments this quarter, up from 18%." That means every comment gets a theme code, the code frame stays stable between waves, and you can click a theme to read the raw verbatims behind it. Manual coding does this well but does not scale past a few hundred comments per wave. Automated coding scales, but only if you can audit it, which means keeping the raw text and the assigned code side by side.
Close the loop
The single strongest driver of value in an NPS program is not the survey design at all. It is whether a detractor hears back from a human within a working day or two. Score-only programs stall because nothing happens after the survey. Programs with a routing rule, an owner and a service-level target for follow-up contact generate retention effects the survey itself never could.
Where to Go Next
Run your numbers with the NPS calculator
Improve the questions around the score: customer satisfaction survey questions
Pick the software layer: voice of customer tools in 2026
Get more out of the open responses: how to get 3x more insights using AI surveys
Frequently Asked Questions
What is a good Net Promoter Score?
There is no single number that qualifies as good across the board. What counts as strong depends on your industry and the countries your respondents live in, because categories differ in how recommendable they are and because rating-scale habits differ by culture. The reliable benchmark is your own score over time, collected the same way each wave.
Is NPS a percentage?
No. It is calculated from two percentages but is reported as a plain number between -100 and +100. Writing "our NPS is 42%" is incorrect. Write "our NPS is 42" or "+42".
Do passives count in the NPS calculation?
Yes and no. Passives are not added or subtracted in the numerator, but they stay in the denominator when you work out the promoter and detractor percentages. Dropping them from the base is the most common calculation mistake and inflates the score.
How many responses do I need for a reliable NPS?
More than most teams assume. NPS is a difference between two proportions, so it carries more sampling error than a single proportion at the same base. Small bases produce large swings that look like real movement. Set a minimum base per reporting cut, and never report an NPS for a segment with a handful of responses.
How often should I run an NPS survey?
Relational NPS is usually run on a fixed cycle, often quarterly or twice a year, because relationship-level sentiment does not move fast. Transactional NPS runs continuously off events. Running relational surveys more frequently mostly adds noise and survey fatigue rather than insight.
Should I replace NPS with something else?
You do not have to choose. Keep NPS if your organization already reports on it, since breaking a trend line has a real cost. Add the mean of the raw 0-10 scores, which uses your data more efficiently and picks up change earlier, and add coded open responses so you know what is driving the movement.
What is the difference between transactional and relational NPS?
Relational NPS is sent on a schedule to a cross-section of your customer base and asks about the overall relationship. Transactional NPS is triggered by a specific event, such as a support ticket or a delivery, and asks about that experience. They produce different numbers from different samples, so they should be tracked as two separate metrics.
