An NPS survey measures customer loyalty with one question: "How likely are you to recommend us to a friend or colleague?" on a 0 to 10 scale. You calculate the Net Promoter Score by subtracting the percentage of detractors (0 to 6) from the percentage of promoters (9 to 10). The result runs from -100 to +100, and anything above 0 means you have more promoters than detractors.
That is the whole mechanic, and it is why NPS spread so fast: it is one question, one number, and a scale a board can read at a glance. The trouble starts after the number. A score with no follow-up is a vanity metric, and a score collected the wrong way is worse than no score at all because it looks trustworthy. This guide covers the question, the math, honest benchmarks, and the operational habits that make the number mean something.
What is an NPS survey?
An NPS survey is a one-question customer loyalty survey that asks how likely a customer is to recommend your company, product, or service, scored from 0 (not at all likely) to 10 (extremely likely). Developed by Fred Reichheld and Bain & Company in 2003, it groups respondents into promoters, passives, and detractors based on their score, and reports a single number that tracks the balance between your most and least loyal customers over time.
The reason it caught on is not that it is the most accurate loyalty measure. It is that it is the most repeatable. Everyone asks the same question the same way, so the number is comparable across quarters, teams, and even companies. That comparability is the whole product. The moment you start editing the question to sound nicer, you lose the one thing NPS is good for.
The Net Promoter Score question
The standard Net Promoter Score question is: "On a scale of 0 to 10, how likely are you to recommend [company or product] to a friend or colleague?" You then add a single open-text follow-up: "What is the main reason for your score?" The rating question produces the number; the open-text question produces the reasons, and the reasons are where all the useful work is.
A few rules keep the question comparable and honest:
- Keep the 0 to 10 scale. Not 1 to 10, not 1 to 5. The category boundaries (detractor, passive, promoter) only work on the standard eleven-point scale.
- Do not lead the respondent. "How much do you love us?" is not an NPS question. The value is a neutral prompt that lets unhappy customers say so.
- Ask about the right thing. Recommending "the company" and recommending "the onboarding you just finished" are different questions. Pick one deliberately and keep it consistent (see relationship versus transactional below).
- Always include the open-text follow-up. The number tells you the score moved. The comments tell you why, and why is the only actionable part.
How to calculate NPS
To calculate NPS, sort every response into three groups, convert two of them to percentages of total responses, and subtract. Promoters are scores of 9 and 10, passives are 7 and 8, and detractors are 0 through 6. The formula is: NPS = % promoters minus % detractors. Passives count toward your total response base but not toward the score itself.
Here is a worked example. Say you collect 200 responses: 110 promoters, 50 passives, and 40 detractors.
| Group | Score range | Count | % of total |
|---|---|---|---|
| Promoters | 9 to 10 | 110 | 55% |
| Passives | 7 to 8 | 50 | 25% |
| Detractors | 0 to 6 | 40 | 20% |
NPS = 55% promoters minus 20% detractors = +35. Notice the passives (25 percent) never enter the subtraction. They pull your promoter percentage down by taking up response share, which is exactly why a wall of "pretty happy" 7s and 8s can hold your score flat even when nobody is actively unhappy. Moving passives up to promoters is often the fastest way to raise a score, and it is usually a smaller lift than rescuing detractors.
What is a good NPS score?
Any NPS above 0 means you have more promoters than detractors, a score above +30 is generally considered good, above +50 is excellent, and above +70 is world-class. But raw benchmarks are misleading without industry context: software and professional services often run higher, while industries with unavoidable friction (utilities, internet providers, insurance) run much lower, and a +20 in one sector can be stronger than a +45 in another.
Two things matter far more than hitting a benchmark number:
- Your own trend. The most useful comparison is against yourself last quarter, with the same question, same audience, and same timing. A score moving from +18 to +26 tells you more than knowing the industry average is +31.
- The distribution behind the number. Two teams can both post +30 while one has almost no detractors and the other has a large promoter base masking a growing detractor problem. Always read the three groups, not just the net.
Relationship NPS vs transactional NPS
There are two ways to run the survey, and mixing them quietly ruins the number. A relationship NPS asks about the overall relationship on a fixed cadence (often quarterly), independent of any single interaction. A transactional NPS fires right after a specific event (an onboarding completed, a support ticket closed, a renewal) and measures that moment.
| Aspect | Relationship NPS | Transactional NPS |
|---|---|---|
| When it fires | Fixed cadence (e.g. quarterly) | Right after a key interaction |
| What it measures | Overall loyalty to the company | Quality of one specific experience |
| Best for | Board-level trend, benchmarking | Fixing a specific process |
| Risk | Slow to reveal the cause of a drop | Skewed by recency of the event |
Run both if you can, but report them separately. A relationship score tells leadership whether loyalty is trending up. Transactional scores tell the onboarding team, the billing team, and the support team whether their specific handoffs are helping or hurting. The transactional version is where NPS connects to real operational fixes, because it points at a single process you can go change.
Sample NPS survey questions and follow-ups
The core rating question stays fixed, but your open-text follow-up should adapt to the score so you learn the right thing from each group. Good practice is to branch the follow-up:
- Rating question (everyone): "How likely are you to recommend [company] to a friend or colleague, from 0 to 10?"
- Detractors (0 to 6): "We are sorry we fell short. What would have made this a better experience?"
- Passives (7 to 8): "Thank you. What one thing would have made this a 9 or 10?"
- Promoters (9 to 10): "Great to hear. What did we do that you would tell a colleague about?"
Keep it to the rating plus one follow-up. Research on survey design consistently shows response rates fall as you add questions, and an NPS survey earns its keep precisely because it is short. If you want to measure more than loyalty, that is a job for a broader instrument, not for bolting five extra questions onto the NPS. The detailed question bank in our guide to customer satisfaction survey questions and examples covers where those additional questions belong.
How to run an NPS survey so the number means something
The survey is easy. Running it so the score is trustworthy and actionable is the hard part, and it is almost entirely operational.
- Send it at a consistent moment. For relationship NPS, pick a fixed cadence and hold it. For transactional NPS, tie it to the same trigger every time. Inconsistent timing makes quarter-to-quarter comparison meaningless.
- Do not survey the same people to death. Set a suppression window so a customer is not asked again for weeks or months. Over-surveying tanks response rates and biases who bothers to reply.
- Watch your response rate as closely as the score. A rising score on a falling response rate usually means only your fans are still answering. The score is drifting up because the unhappy stopped replying, not because they got happy.
- Route detractor comments to an owner within a day. A detractor who wrote a specific complaint is a churn risk you can still save, but only if someone reads it and acts before the feeling hardens.
- Report the three groups, not just the net. Give leadership the promoter, passive, and detractor counts alongside the single number so the story behind a flat score is visible.
NPS, CSAT, and CES: which loyalty metric to use
NPS is one of three common experience metrics, and they answer different questions. NPS measures loyalty and likelihood to recommend over the relationship. CSAT (Customer Satisfaction) measures how satisfied someone was with a specific interaction. CES (Customer Effort Score) measures how hard the customer had to work to get something done. None replaces the others.
| Metric | Question it asks | Best for | Scale |
|---|---|---|---|
| NPS | Would you recommend us? | Overall loyalty, trend, benchmarking | 0 to 10, reported -100 to +100 |
| CSAT | How satisfied were you with this? | Satisfaction with a single interaction | Typically 1 to 5 |
| CES | How easy was it to get this done? | Friction in a specific task or process | Typically 1 to 7 |
A practical setup is to use NPS as the relationship-level trend line, CSAT to check specific touchpoints, and CES to find the friction that is quietly creating detractors. If your NPS is falling and you do not know why, a well-placed Customer Effort Score survey at the process level often surfaces the cause before the NPS comment box does, because effort is where loyalty leaks first.
Why does my NPS score not improve even when I fix things?
The most common reason a score stays flat despite real fixes is that NPS is a lagging, relationship-level metric, so improvements to a single process show up slowly and diluted. If you fixed onboarding but bill customers late, the relationship score nets the two out. NPS moves when the whole experience improves, not when one team ships one fix, which is exactly why transactional NPS and process metrics are better for proving a specific change worked.
The second reason is that you collected the score but never closed the loop. Customers who take the survey and see nothing change stop believing the survey matters, response rates decay, and the number stops reflecting reality. Closing the loop is not optional politeness; it is what keeps the instrument accurate. Our guide to the customer feedback loop walks through the acknowledge, route, fix, and report-back sequence that keeps a score honest over time.
Where NPS fits in a wider program
NPS is a thermometer, not a treatment. It tells you loyalty is up or down; it does not tell you what to do, and it cannot fix anything on its own. That is why the strongest programs treat NPS as one input inside a structured voice of customer program that also listens to support tickets, churn reasons, and interviews, then routes what it hears to the teams who own the work.
And the work is almost always operational. A detractor rarely writes "your NPS survey was bad." They write that their first invoice was wrong, that onboarding took two weeks, that they had to explain the same problem to three people. Those are back-office process failures wearing a sentiment score, which is the throughline in our foundational piece on why customer experience is won in the back office. The score is where you notice the problem. The operation underneath it is the only place you can fix it.
Start with one survey, one trigger, one owner
You do not need a platform or a survey program to start. Pick one moment that matters (post-onboarding is a strong first choice), ask the single recommend question plus one open-text follow-up, and give one person the job of reading detractor comments and acting on them each week. Report the three groups and the net. That is a complete, honest NPS loop. Once it earns trust and shows a stable trend, widen it. A small survey someone acts on beats a company-wide rollout nobody follows up.