What is Net Promoter Score?
Net Promoter Score (NPS) is a customer loyalty metric derived from a single question: how likely a customer is to recommend a company to a friend or colleague, answered on a zero-to-ten scale. Responses fall into three bands, and the score is the percentage in the top band minus the percentage in the bottom band.
Published in 2003, NPS spread because it compresses a relationship into one number a board can track quarter over quarter. It runs from -100 to +100. Because middle responses sit in the denominator while contributing nothing to the subtraction, most companies land far below the theoretical maximum.
How the Net Promoter Score formula works
An NPS program runs in four steps: sample, ask, band, subtract. Sampling decides who is surveyed and when, usually a rolling slice of active customers for relational NPS, or a trigger fired after a closed ticket for transactional NPS, which puts it in the same slot as Customer Satisfaction Score. The ask is deliberately short: the rating plus one open text field asking why, and that free text is what later feeds conversational analytics alongside chat and call transcripts.
Banding is fixed. Nine and ten are promoters, seven and eight are passives, zero through six are detractors. The formula is promoters as a percentage of all responses minus detractors as a percentage of all responses.
Work it through. Of 400 responses, 220 are promoters (55%), 100 are passives (25%), and 80 are detractors (20%). The score is 55 minus 20, or 35. If 40 of those promoters slip to passive, promoters fall to 45%, detractors hold at 20%, and the score drops to 25 without a single new complaint.
Nothing in that arithmetic covers what happens next. The follow-up that turns a detractor's comment into a shipped fix belongs to a customer feedback loop, which the score itself never enforces.
What counts and what does not
Promoters (9-10): Ratings of nine and ten count fully toward the positive side of the subtraction and are treated as willing advocates, whether or not they ever refer anyone.
Passives (7-8): Sevens and eights enter the denominator and stay out of the subtraction, so a wave of them drags the score down invisibly.
Detractors (0-6): The detractor band is wide, so a six who is mildly disappointed counts exactly as heavily as a zero who is cancelling.
Non-responders: People who ignore the survey never enter the calculation at all, even though they are almost always the majority of the sample.
Mixed cadences: Relational and transactional responses answer different questions, so pooling them into one series produces a number nobody can act on.
Net Promoter Score vs CSAT vs CES vs customer health score
Teams run all four and then argue about which one belongs in the board deck. Customer Satisfaction Score rates one interaction in the minutes after it ends. Customer Effort Score rates how much work the customer had to do to get that interaction resolved. Customer health score blends product usage, support history, and billing signals into a renewal prediction owned by an account team. Net Promoter Score asks about the whole relationship and about a future act, a recommendation, which makes it the slowest of the four to move and the hardest to attribute to any one ticket.
What it counts | What it misses | Typical benchmark | |
|---|---|---|---|
Net Promoter Score | Stated likelihood of recommending the company, promoters minus detractors | The reason behind the rating, and everyone who ignored the survey | SaaS averages sit near 30-40, retail closer to 40-50, telecom often below 20 |
CSAT | Satisfaction with one specific interaction, rated immediately afterwards | Whether that satisfaction survives until the renewal date | Reported as percent positive, and varies sharply by channel and by prompt wording |
CES | How much work the customer did before the issue was resolved | Feelings about price, product quality, or the brand overall | Reported on a seven-point agreement scale, read as a distribution |
Customer health score | Usage, support, and billing signals combined into a renewal prediction | Anything a customer feels strongly but never does in-product | Defined per vendor and per segment, so no cross-industry norm exists |
If the question is whether last night's chat worked, ask CSAT or CES within the hour. If the question is whether loyalty is compounding across a whole book of customers, use NPS, and accept that it will lag whatever your team fixed last week.
Why Net Promoter Score matters for customer experience
Without a loyalty metric, a support organisation optimises what it can see hourly: queue depth, handle time, and resolution rate. All three can improve while customers quietly decide they would steer a colleague elsewhere, because a fast, correct, and cold resolution still spends goodwill.
The failure mode when NPS is absent is silent. Detractor comments never leave the survey tool, product never hears the same complaint twice in writing, and churn shows up as a surprise in a renewal forecast that looked healthy on operational dashboards.
The tradeoff is real. NPS moves slowly, needs volume before it is stable, and invites gaming the moment it becomes a bonus target, so a team that treats a two-point quarterly shift as a signal will spend the next quarter chasing noise.
How is Net Promoter Score measured over time?
Measured once, NPS is a snapshot with a wide error bar around it. Measured identically every quarter, it becomes a trend, and the trend is the only form of it worth putting in front of a board. Consistency carries the whole discipline: the same sampling frame, the same channel mix, the same question wording, and the same interval between asks. Change any of those and the movement you see is a methodology artefact.
Response bias is the standing threat. People who answer skew toward the delighted and the furious, so the ambivalent middle is underrepresented, and a rising score can simply mean the indifferent stopped replying.
Closing the loop is where the cost sits, and the U.S. Bureau of Labor Statistics puts median pay for customer service representatives at USD 20.59 an hour, or USD 42,830 a year, so a five-minute detractor callback costs roughly USD 1.70 in wages before any tooling.
How AI agents change Net Promoter Score
AI agents change the population being surveyed before they change the score. When an AI agent closes most first-contact tickets, the transactional survey lands mostly on cases no human ever saw, and the relational survey covers customers whose most recent memory of the company is a chat window. Score movement after deployment often reflects that mix shift rather than a change in service quality, which is why the resolving channel has to be tagged on every response.
The second change is on the open text field. Verbatims that once sat unread in a spreadsheet can now be clustered by intent and sentiment at full volume, so the "why" becomes a ranked list of causes. Buyers evaluating platforms on resolution and CSAT outcomes should ask the same of loyalty data.
Faster containment also compresses average resolution time, and speed alone rarely converts a detractor into a promoter.
Implementing a Net Promoter Score program
Start with the sampling frame, because it decides everything downstream. Define who is eligible, how often a given account can be asked, and whether the score is relational, transactional, or both reported separately.
Integration surface is next. The response has to write back to the CRM and the help desk record so a detractor's rating is visible to whoever picks up their next ticket, and so verbatims can be joined to product usage.
Governance is the axis teams skip. Someone must own detractor follow-up with a named response window, and someone must decide who can export verbatims, since free text routinely contains account numbers and health or payment details. Regulated buyers ask how those verbatims are stored, who can read them, and whether deletion requests reach the survey tool, which is where GDPR and SOC 2 Type II questionnaires land.
The constraint that bites hardest is survey fatigue: NPS, CSAT, and product research all draw on the same inboxes, and without a shared frequency cap the response rate collapses first among the customers you most need to hear from.
Net Promoter Score and customer success metrics
NPS rarely earns its keep alone. Pair it with customer health score, which combines usage, support, and engagement signals into a renewal prediction, and a low rating from a high-usage account becomes an urgent, specific conversation instead of a data point.
Pair it with Customer Effort Score and you get the mechanism behind the rating: effort explains why an account that ships tickets weekly has stopped recommending you, which the recommendation question alone never reveals.
What does Net Promoter Score mean in plain terms?
NPS stands for Net Promoter Score, and the "net" is the accounting sense of the word. Think of it as a room full of your customers where you count the people who would bring a friend, subtract the people who would warn one off, and ignore everyone who shrugs.
Here is what that hides. If ten customers rate you a ten and ten rate you a two, the score is zero. If all twenty rate you an eight, the score is also zero. One of those companies has passionate fans and a serious problem; the other has nobody who cares much either way, and the single number cannot tell them apart.
The tradeoff is compression. You gain a figure any executive can read in a second, and you lose the distribution that would have told you what to fix, which is why the comment field matters more than the digit.
Common Net Promoter Score mistakes
Making the score a personal target for the people who collect it is the first failure. Once a bonus depends on it, the sampling frame becomes something staff influence: happy customers get surveyed, unhappy ones get quietly skipped, and the number rises while nothing about the service changes.
Pooling relational and transactional responses into one series is the second. The two ask about different time horizons, so the blended figure moves whenever the ticket mix moves, and no one can trace a drop to a cause.
Reading only the digit is the third. The rating tells you the temperature and the verbatim tells you the reason, so a program that never routes comments to an owner has built a thermometer with no thermostat attached.
The fourth is designing detractor follow-up as a fully automated sequence. A one-star rating on a billing dispute is exactly the case that needs a person, which is an argument for deliberate human fallback in AI chat on any recovery path.
What is a good Net Promoter Score?
A good Net Promoter Score depends entirely on the industry and the sampling method behind it. SaaS averages sit around 30-40, retail closer to 40-50, and telecom often below 20. Anything positive means promoters outnumber detractors. Your own trend across identical quarterly measurements tells you more than any cross-industry comparison will.
How do you calculate NPS from survey responses?
NPS is calculated by banding every response, then subtracting. Ratings of nine and ten are promoters, seven and eight are passives, zero through six are detractors. Take promoters as a percentage of all responses, subtract detractors as a percentage of all responses, and the result is a whole number between -100 and +100.
What is the difference between NPS and CSAT?
NPS and CSAT measure different time horizons. CSAT asks how satisfied someone was with one specific interaction, captured minutes after it ends, and reads as percent positive. NPS asks about the whole relationship through a hypothetical recommendation, and reports a net figure. CSAT reacts to a ticket; NPS reacts to a year of them.
Is transactional NPS different from relational NPS?
Transactional NPS and relational NPS use the same question and answer different things. Transactional fires after a specific event such as a resolved ticket or a delivery, so it reflects that episode. Relational goes to a sample of the customer base on a fixed cadence, independent of recent activity. Report them as separate series.
Why are passives excluded from the NPS formula?
Passives, the sevens and eights, are counted in the denominator while contributing nothing to the subtraction. The original design treats them as satisfied customers who are still available to a competitor, so they neither help nor hurt. The practical effect is that a large passive population dilutes the score and drags it toward zero.
How often should you run an NPS survey?
NPS surveys are typically relational on a quarterly or twice-yearly cadence, and transactional whenever a defined event closes. Frequency caps matter more than the interval itself: if the same customer receives loyalty, satisfaction, and research surveys in one month, response rates fall and the remaining sample skews toward your most extreme opinions.