Quick answer
Net Promoter Score is calculated from a 0-to-10 likelihood-to-recommend question. Respondents scoring 9 or 10 are promoters, 7 or 8 are passives and 0 through 6 are detractors. NPS equals the percentage of promoters minus the percentage of detractors, producing a value from -100 to +100. The metric was popularized by Frederick Reichheld's 2003 Harvard Business Review article. It can be a useful high-level relationship signal, but academic research has challenged claims that it is universally superior or a reliable single predictor of growth. Teams should define population and timing, report sample and uncertainty, diagnose reasons, connect to behaviour and economics, and never manage only the arithmetic.
What is Net Promoter Score?
NPS is a survey-based recommendation metric popularized by Frederick Reichheld in the 2003 Harvard Business Review article The One Number You Need to Grow. It asks how likely a customer is to recommend a company, product or service, conventionally on a zero-to-ten scale.
Scores of nine and ten are promoters, seven and eight are passives, and zero through six are detractors. NPS is the promoter percentage minus detractor percentage. Passives remain in the denominator but contribute to neither side of the subtraction.
The result summarizes the distribution under those thresholds. It is not the average score, the percentage who actually recommended, or a direct measure of retention, profitability or growth.
NPS formula and worked calculation
Suppose 200 valid respondents include 100 promoters, 50 passives and 50 detractors. Promoters are 50 percent and detractors 25 percent, so NPS is 25. Report the counts and base because the same score can arise from different response distributions.
The theoretical range is -100 when all respondents are detractors to +100 when all are promoters. An NPS of zero means promoter and detractor shares are equal, not that customers are neutral.
Sampling error matters. A point estimate from 80 respondents is less precise than one from 8,000, and repeated customers or clustered accounts require appropriate analysis. Report intervals and avoid ranking tiny differences as fact.
Define
Specify the relationship, population, touchpoint and decision the survey will inform.
- Who can reasonably recommend?
- When should they answer?
Sample
Design contact and weighting to reduce coverage, nonresponse and timing bias.
- Who is missing?
- Are frequent users overrepresented?
Calculate
Apply the standard categories and report counts, percentages and uncertainty.
- Is the denominator clear?
- Is change distinguishable from noise?
Diagnose
Combine reasons, operations and behaviour to locate the customer problem.
- What experience drove the rating?
- What alternative explains movement?
Act
Improve the underlying experience and evaluate outcomes rather than coach the score.
- Which mechanism changes?
- Did behaviour and value improve?
Design the NPS survey
Define whether the question concerns an overall relationship or a recent episode. Relationship surveys sample the ongoing customer base; transactional surveys follow an interaction. Mixing them changes population, timing and meaning.
Use neutral wording, accessible language and an appropriate channel. Prevent duplicate over-surveying, respect consent and avoid asking employees to select happy customers. Record enough context to assess coverage without collecting unnecessary personal data.
Add an open reason or focused diagnostic question, but do not make a long questionnaire mandatory after the rating. Follow-up should help explain experience, not pressure respondents to defend their score.
Sampling, response and comparability
Create a sampling frame that represents the intended population. Excluding churned, unresolved, offline or low-frequency customers can inflate the score. Survey timing can also overrepresent dramatic episodes.
Monitor invite, delivery and response by relevant cohort, product, tenure, market, language and channel. Weighting may address known selection differences, but it cannot fully repair missing groups with unobserved experience.
For trends, keep wording, population, mode and timing stable or document breaks. A process change can move the number without any customer experience change.
Does NPS predict growth?
Reichheld's original article presented recommendation as a strong growth indicator in selected settings. The clarity of the claim contributed to rapid managerial adoption.
Keiningham, Cooil, Andreassen and Aksoy used longitudinal data from 21 firms and more than 15,500 interviews and did not replicate clear superiority over other customer metrics in the studied industries. Later research continues to find context-dependent results.
The defensible conclusion is not that NPS never matters. Willingness to recommend can contain useful relationship information, but a universal single-metric causal claim is unsupported. Validate it against actual outcomes in the relevant business.
Move from score to root cause
Analyze the response distribution and reasons, not only the net. A stable NPS can hide more promoters and more detractors, while one segment's improvement can offset another's decline.
Link survey data carefully to product, service, price and journey evidence. Themes require coding quality and sufficient context. A comment explains one respondent's experience, not automatically the population.
Close the loop on severe or recoverable issues when permission and capacity allow. Separate service recovery from research so respondents do not feel that a critical answer creates unwanted sales contact.
Worked example: diagnosing a repair-service NPS decline
Mendway first finds a coverage break, then combines comments and operational data to identify status uncertainty during parts delays. The response is to improve the experience, not to selectively increase promoter invitations.
The broader scorecard prevents false success. NPS improvement matters more when repair time, effort, repeat failure and trust also improve, and an experiment or stronger design supports the intervention claim.
Mendway is a fictional repair service for commuter bags. Its relationship NPS falls from 42 to 29 in one quarter, and managers propose asking support agents to remind satisfied customers to complete the survey.
The survey covers customers with a completed repair and is sent after delivery. The team separates relationship NPS from a repair-episode question and records the actual user and payer roles.
Response by repair type, resolution time, channel and language is checked. A recent switch in invitations had excluded customers whose cases required manual follow-up.
The team reports promoter, passive and detractor counts, response rate and confidence intervals. It avoids treating every point change as operationally real.
Comments, case data and interviews connect dissatisfaction to parts-delay communication and repeated requests for status, not frontline friendliness alone.
Mendway improves inventory visibility and proactive status updates, then tracks repair completion, effort, complaints, repeat use and later NPS. Agents are not rewarded for soliciting only likely promoters.
Mendway and scores are hypothetical. NPS benchmarks and statistical interpretation depend on survey design, population and context.
NPS governance and incentives
Maintain a survey specification with owner, population, trigger, wording, channel, sampling, exclusions, calculation, weighting and reporting. Version every change and preserve raw category counts.
Do not tie frontline compensation mechanically to the score. Employees may coach customers, delay difficult cases or suppress invitations. Reward root-cause resolution and customer outcomes through a balanced system.
Benchmark cautiously. Competitors may use different populations and modes, and published industry numbers may be vendor samples. Internal consistent trends and behavioural validation are usually more actionable.
NPS limitations
Recommendation relevance varies by category and relationship. A person may value a sensitive, routine or monopolistic service without discussing it. Cultural and scale-use differences can affect response thresholds.
The categorization discards distance within groups: zero and six are both detractors, while six and seven fall on opposite sides. The simplicity is useful for communication but can reduce statistical information.
NPS is attitudinal intent. Actual referral, retention and value should be measured directly where possible. Survey movement is not causal evidence about a campaign or product change.
NPS checklist
Use this checklist before publishing or acting on an NPS result.
- Relationship or episode is specified
- Population and recommendation relevance are clear
- Sampling frame includes difficult cases
- Wording, scale and timing are stable
- Promoter, passive and detractor counts are reported
- Sample size and uncertainty are visible
- Response and coverage bias are assessed
- Reasons are coded with quality controls
- Operational and behavioural outcomes are linked
- Score coaching is prevented
- Benchmarks use comparable methods
- Causal claims match the evaluation design
NPS is a conversation starter and trend signal. It becomes dangerous when simplicity is mistaken for certainty, diagnosis or proof of growth.
Frequently asked questions
How is Net Promoter Score calculated?
Subtract the percentage of respondents scoring zero through six from the percentage scoring nine or ten. Respondents scoring seven or eight remain in the denominator but add to neither percentage.
What is a good NPS?
There is no universal good score. Population, category, market, survey mode and timing differ. Use a consistent internal trend, uncertainty and links to customer behaviour rather than an unverified benchmark.
Do passives count in NPS?
They count in the total respondent denominator but are neither added as promoters nor subtracted as detractors.
Does NPS prove customer loyalty or growth?
No. It measures stated recommendation intent under a survey design. Research has challenged universal superiority and growth claims, so validate against retention, referral and financial outcomes.
How often should NPS be measured?
Use a cadence suited to the decision and natural relationship cycle. Avoid over-surveying and keep population, trigger and method stable enough for comparison.
Sources and further reading
- Harvard Business Review: The One Number You Need to Grow ↗Reichheld's original 2003 article popularizing the recommendation metric
- Journal of Marketing: Longitudinal Examination of Net Promoter and Growth ↗Independent replication challenging claims of clear metric superiority
- CiNii Research: Longitudinal Examination of Net Promoter ↗Accessible bibliographic record and abstract for the 21-firm study
- Journal of Business Research: Systematic Evaluation of NPS Calculation ↗Longitudinal comparison of NPS and alternative recommendation calculations