Sopact is a technology based social enterprise committed to helping organizations measure impact by directly involving their stakeholders.
Copyright 2015-2026 © sopact. All rights reserved.
The NPS formula is one line of arithmetic: % promoters minus % detractors. The calculation is settled. Measuring NPS is the four strands, on one record.
The NPS formula is the percentage of promoters minus the percentage of detractors. One line of arithmetic, settled in 1993. The same line cannot tell two very different programs apart, cannot say what changed, and cannot name what to fix. Sopact measures NPS by reading every verbatim that arrives with the score, against the same contact's prior wave — so the number arrives with the reason underneath.
There is no version, no asterisk, no proprietary extension. The formula is one subtraction, and it has been the same since 1993. Most of what teams call "how to calculate NPS" is actually three small things: the buckets, the percentages, and the subtraction.
NPS = % Promoters − % Detractors
Passives are excluded from the subtraction. They count in the denominator (total responses) but not in either percentage of the numerator.
Loyal advocates. They recommend, refer, and stay. Their verbatim usually names what specifically earned the high score.
Satisfied but unenthusiastic. Out of the formula by design — the research found their behavior closer to detractors than promoters.
Unhappy customers. They churn, they tell others, they cost the program more than the score suggests. Their verbatim is the most load-bearing data in the entire NPS program.
The buckets are not arbitrary. The 9-10 / 7-8 / 0-6 cutoffs reflect the original Bain/Reichheld research that found these three groups behave very differently — in retention, in referrals, in lifetime value. A vendor that defaults to a different cutoff is not running NPS; it is running something else.
A B2B SaaS company sends a quarterly relational NPS survey to two thousand customers. Twelve hundred respond. The calculation is below, step by step, with the exact arithmetic any team can replicate in a spreadsheet.
Every respondent is classified into one of three buckets by their score. The percentages of promoters and detractors are calculated against the total response count. Passives are excluded from the subtraction but counted in the denominator.
A spreadsheet can do this in three formulas. A modern survey tool does it on a dashboard. This is the part that is settled. Everything interesting about measuring NPS begins after the result.
For twenty years, "how to measure NPS" meant the four steps in the worked example above. A team that could get to step 5 reliably each quarter felt like they were measuring NPS. The vendors made dashboards. The dashboards reported the number. The number was the deliverable.
Two things changed. First, the calculation became trivial — any survey tool, any analytics platform, any spreadsheet does step 5 instantly. The arithmetic stopped being the bottleneck twenty years ago and the AI era only made it more obvious.
Second, the standard of what counts as "measuring NPS" has moved. A number on a dashboard no longer qualifies. The work moved to reading the verbatim that arrives with the score, on the same contact across waves, against any other record attached to that contact — so the number arrives with the reason underneath.
The score answers what the number is. The verbatim answers why. The trajectory (the same contact across waves) answers since when. The action (what the team did about the verbatim) answers whether the loop ever closed. A program tracking only the score is doing one-quarter of the measurement and reporting it as the whole.
This is the same argument that anchors /use-case/nps-analysis, the analytical pillar of the NPS cluster — expressed here through the methodology frame. The pillar covers what NPS analysis is; this page covers what to actually measure.
Both programs below produce an NPS of 40. The board deck shows the same headline. The customers, the verbatims, and the next quarter's churn risk are very different. The formula does not see that.
A divisive product or program. The promoter base is loyal and growing; the detractor base is real and naming specific failures. The formula nets them to 40, but the verbatims tell two completely different stories. The next quarter's risk is in the detractor verbatim, not the score.
"This used to be the only tool we trusted. The last release broke the integration we depend on. We are evaluating alternatives." — one of 240 detractor verbatims.
A safe, low-risk program. Few enthusiastic advocates but very few critics. The verbatims sound similar to each other — functional satisfaction, no real pain. The formula nets them to 40, but this program's risk is opposite — the passives will leave quietly when a sharper competitor arrives.
"It works. We use it. No complaints, no surprises." — representative of the 600 passive verbatims.
A dashboard that shows 40 next to 40 is doing arithmetic. A measurement is what tells the team these are two different programs. The arithmetic is settled; the measurement is what reads the verbatim, the distribution, and the trajectory together.
The score is one strand. The full measurement is four, each on the same persistent contact ID. A program tracking only the score is reporting the arithmetic and calling it the measurement.
The arithmetic. A number on a -100 to 100 scale. Quantitative, comparable across waves, useful for trends. Settled in 1993.
The customer's open-ended answer. Qualitative, names the reason behind the score. The only strand the team can act on directly.
The same contact's score and verbatim across prior waves. A 4 from a former 9 means something different from a 4 from a new contact. Visible only with a persistent contact ID.
What the team did about the verbatim, logged on the same record. Did the detractor get a follow-up call? Did the score move next wave? This is whether the loop ever closed.
Strand 01 is the part the formula produces. Strands 02-04 are the part the team has to build the discipline around. The first is twenty seconds in a spreadsheet; the others take a workflow that reads every response on arrival, on one record, against the prior wave.
Five mistakes show up in almost every program that "measures NPS" but cannot tell what to do with the result. None of them are about the arithmetic.
Under a hundred responses the score swings by ten or more points between waves on noise alone. The team reads movement that is not there. Confidence intervals widen. Below 100 responses, the score is not a measurement — it is a guess with a decimal point.
An NPS that "moved" from 38 to 42 might be the same population behaving differently or three times as many respondents in a different distribution. The arithmetic does not tell which. Wave-over-wave comparison without acknowledging response-count variance produces false trends and false alarms.
The most common mistake by an order of magnitude. The score goes on the dashboard. The verbatim goes in an export folder. The team reports movement they cannot explain because the explanation was in the column nobody opened. A score without the verbatim is one strand of a four-strand measurement.
Some vendors default to a 1-10 NPS scale. The cutoffs do not map cleanly to the validated 9-10 / 7-8 / 0-6 buckets. The result looks like NPS, behaves slightly differently, and cannot be compared to any external benchmark. Check the scale. Change it to 0-10 if the default is wrong.
The most expensive mistake. The same customer's Q4 response has no relationship to their Q3 response. Trajectory disappears. Drift is invisible. A 9-to-4 trajectory, the most important signal in an NPS program, reads identically to a one-off 4 from a new contact. Without a persistent contact ID, the measurement is wave-by-wave even when the team says it is longitudinal.
None of the five is a calculation error. The formula was right every quarter. The measurement was wrong every quarter. That is the gap measuring NPS in 2026 has to close.
This page covers the calculation and what surrounds it. The cluster covers the analytical pillar, the informational hub, and the commercial sub-hub. Same locked argument, different reading entry.
What NPS analysis means in 2026, the workflow, the AI-era thesis, the longitudinal context. The broader treatment above this page.
Read the pillar →The formula, the worked calculation, the four strands of measurement, and the five common mistakes that have nothing to do with the arithmetic.
This pageThe full feedback signal — score, verbatim, prior history, attached context — and the closed-loop workflow that has to fire around each response.
Read the hub →If you came for the formula, you are in the right room. If you came for the methodology and the AI-era argument, the pillar is the right next read. If you came to compare verbatim-reading tools, the commercial sub-hub is.
Same instrument, same arithmetic. The cost of measuring NPS at strand-01-only is different in each context — and so is the value of measuring all four strands.
A quarterly relational NPS run with a persistent contact ID across waves. The CS lead can see each at-risk account's score and verbatim trajectory before the QBR, not after the churn. The score arrives with the reason. The reason arrives with the prior reason.
A training program with an end-of-program NPS plus a 90-day and 180-day follow-up, run on a persistent participant ID. Cohort movement reads not as a single number but as a pattern across strands — the score, the verbatim, the same participant's six-month follow-up. The funder report writes itself.
A scholarship program with NPS-style feedback after each cycle and a six-month follow-up. Each awardee's score and verbatim sit on their persistent application record alongside the original essay. The change story emerges from the data — the score moved this much, here is the verbatim, here is the awardee's name and original goal.
Your scores, your verbatims, your contacts. Sixty minutes. No demo accounts.
NPS is calculated by subtracting the percentage of detractors (respondents who scored 0-6) from the percentage of promoters (respondents who scored 9-10). Passives (7-8) are excluded from the calculation. The result is a number on a -100 to 100 scale. Example: 60 percent promoters minus 20 percent detractors equals an NPS of 40. The arithmetic is the easy part of measuring NPS.
NPS = percent of Promoters − percent of Detractors. Promoters are respondents who score 9 or 10. Detractors are respondents who score 0 through 6. Passives (7 or 8) do not enter the formula. The result can range from -100 (every respondent is a detractor) to +100 (every respondent is a promoter).
Promoters scored 9 or 10 on the 0 to 10 question. Passives scored 7 or 8. Detractors scored 0 through 6. The cutoffs are not arbitrary — they reflect the original Bain/Reichheld research that found 9-10 customers behave very differently (referrals, repeat purchase, advocacy) than 7-8 customers, who behave more like detractors than promoters in practice.
There is no single answer because benchmarks vary by industry, audience, and program type. Any positive score means more promoters than detractors. A score above 50 is generally considered strong. A score above 70 is unusual. But comparing your score to an industry benchmark is less useful than comparing it to your own prior wave — the trajectory matters more than the absolute number.
Yes. NPS ranges from -100 (every respondent is a detractor) to +100 (every respondent is a promoter). A negative NPS means more detractors than promoters — the program has more critics than advocates. A negative score is not automatically a failure mode; it is a signal to read the verbatims and understand why.
Under 100 responses the score swings significantly between waves and confidence intervals get wide. 200 to 400 responses give a stable enough read for quarterly tracking. Above 1,000 the calculation is statistically stable, but the same waves can hide a churning detractor base offset by a few new promoters. Sample size answers a precision question, not a what-to-fix question.
The standard NPS scale is 0 to 10 (eleven points). Some tools offer a 1 to 10 scale, but it is not the original instrument and the cutoffs do not map cleanly. The 0 to 10 scale is what the Bain/Reichheld research validated. Use it. If a vendor offers a 1 to 10 scale by default, change the setting.
Three cadences are common. Relational NPS is collected on a fixed schedule (usually quarterly) and reads the relationship overall. Transactional NPS fires after a specific event (a support ticket, a release, an onboarding) and reads that event. Continuous NPS samples a rolling subset of customers each week. Quarterly relational is the most common; cadence is a discipline question, not a formula question.
Because the score is an arithmetic compression. A program with 70 percent promoters, 10 percent passives, and 20 percent detractors produces NPS 50. A program with 50 percent promoters, 50 percent passives, and 0 percent detractors also produces NPS 50. These are very different programs. The score alone tells you nothing about which one is yours; the verbatim and the distribution do.
Use a persistent contact ID. Without one, every quarterly survey is a fresh export and the same customer's score this quarter has no relationship to their score last quarter. With a persistent ID, the team can see each customer's trajectory — a 9 dropping to a 4 is a different story than a 4 from a new contact. The trajectory is the part the formula does not produce.
Five common mistakes. One, calculating the score on a sample too small to be stable. Two, comparing waves with different sample sizes without acknowledging the variance. Three, treating the score as the measurement and ignoring the verbatim. Four, using a 1-10 scale instead of the standard 0-10. Five, running each wave as a fresh export with no persistent contact ID, so the trajectory disappears. The formula is rarely the failure mode; the discipline around it usually is.
Four things. The score (the arithmetic). The verbatim (the reason). The trajectory (the same contact's score and verbatim across prior waves). The action (whether the loop closed after the response — what the team did, whether the customer's score moved next wave). Tracking only the score is using the smallest part of the measurement. The other three are where the work that improves the score actually happens. See /use-case/nps-analysis for the broader treatment.
The arithmetic is settled. The cluster covers the methodology, the informational hub, and the commercial sub-hub.
Your scores, your verbatims, your contacts. Sixty minutes. We walk through the calculation, then through what the calculation does not say — the verbatim reading, the trajectory per contact, the action that should have followed each detractor response. No demo accounts. No slideware. Your own data, measured live.
No slideware. No demo accounts. Your own data, measured live.