How do you measure NPS?
Measure Net Promoter Score by asking a recommendation question on a 0–10 scale, grouping the valid ratings and subtracting the detractor percentage from the promoter percentage. To make the result useful, define the audience and timing before you send the survey, then plan how the team will review feedback and respond.
The calculation needs ratings, not written explanations. An optional follow-up often adds useful context, but a missing comment does not make a valid rating unusable. Likewise, personal identity is not required to calculate NPS. Decide whether you need individual follow-up or matching across time as a separate design choice.
This guide is a setup checklist for a first or recurring measurement. Use NPS analysis for interpreting groups, comments and changes after collection.
1. Decide what experience you are measuring
Write down whose perspective you need and what the result will support. An overall relationship check and a survey sent after a support interaction have different contexts. Keep them distinguishable in collection and reporting.
Scroll horizontally to see all columns →
| Measurement purpose | Example audience | Design question |
|---|---|---|
| Overall customer relationship | A defined set of current customers | Have they had enough experience to give a meaningful response? |
| Feedback after a service event | People who completed the relevant interaction | Does the question refer clearly to the brand or to that specific event? |
| B2B account understanding | Relevant roles within customer accounts | Will you report individual perspectives, account summaries or both? |
If your immediate question is whether a task was easy or a request was resolved, a task-specific measure may be more direct. Do not make NPS carry every measurement purpose. Use other evidence when the decision requires it.
2. Use a clear recommendation question
A common formulation is: “How likely are you to recommend [organization or service] to a friend or colleague?” Show every value from 0 to 10, with endpoints explaining that 0 is not at all likely and 10 is extremely likely.
Check whether recommending the service is meaningful for the audience. Some services are not freely chosen or are private in a way that makes recommendation awkward. If you adapt the wording, document it and be careful with external comparisons.
Add a focused optional question such as “What most influenced your rating?” A second question may be useful when it supports a different action, but avoid turning a short check-in into an unplanned long questionnaire. See NPS question examples and follow-ups.
Do not describe higher scores as the answer staff need or ask only apparently satisfied customers. Both practices undermine the purpose of measuring the experience.
3. Define who can respond
Create an invitation rule before looking at results. Specify the eligible population, exclusions and unit of response. For example, “current customers with at least one completed service interaction in the last three months” is clearer than “people in our mailing list.”
Decide whether one person can answer more than once in a period. A relationship survey usually needs a different duplicate rule from an interaction survey. Preserve the rule in the reporting notes rather than removing responses inconsistently after seeing their ratings.
For a customer account with several respondents, retain the respondent role where appropriate. An administrator and an everyday user may have different experiences. Explain whether the report counts responses or accounts. Do not silently give larger accounts more weight simply because more people answered.
4. Choose timing and a sustainable cadence
There is no universal best frequency. Match the cadence to the pace of the experience and the team’s ability to act. A short onboarding journey and a yearly membership relationship may need different arrangements.
For event-triggered feedback, define the trigger and delay. For recurring relationship feedback, define the eligible population and reporting window. Limit repeated invitations so that people with frequent interactions are not overburdened.
Hold important design choices steady when you want a trend, and record changes when they are necessary. Consistent wording does not guarantee a comparable sample, but it removes one avoidable source of difference.
5. Choose anonymous or identified collection deliberately
An anonymous survey can measure a group’s NPS and keep each comment with its rating under a submission record. It cannot support direct follow-up with that respondent unless they separately provide suitable contact details.
Identified collection can support authorized customer follow-up and matching across periods. Explain the purpose, use appropriate access and avoid collecting more personal context than the task requires. A unique invitation link may help manage responses, but it is not a guarantee of anonymity.
Separate the views needed by different audiences. A leadership report may need category counts, themes and actions; a service owner may need specific contact context. Verify the privacy promises in the questionnaire against the actual configuration.
6. Calculate NPS correctly
Under the standard classification, promoters rate 9–10, passives 7–8 and detractors 0–6. Calculate each percentage using all valid ratings, then subtract detractors from promoters. Bain’s measurement guide provides the standard formula.
Worked example: 250 people are invited, 100 provide a valid rating and 60 of those also leave a comment. Among the 100 ratings are 50 promoters, 30 passives and 20 detractors.
- Promoters: 50 ÷ 100 = 50%.
- Detractors: 20 ÷ 100 = 20%.
- NPS: 50 − 20 = +30.
- Comment coverage: 60 of 100 valid ratings have a comment.
Use 100 as the scoring denominator, not the 250 invitations or the 60 comments. Include passives in that denominator. Do not code unanswered ratings as zero: zero is a valid rating with a different meaning.
Report the counts with the score. For pooled results, calculate from all valid responses or use the appropriate response-count weighting for nonoverlapping groups. Keep any deliberate account or survey weighting transparent.
7. Test the full collection path
Before launch, submit test responses covering 0, 6, 7, 8, 9 and 10 so category boundaries can be checked. Try a missing comment, a missing rating, a duplicate and an answer on a phone. Confirm the question and scale are accessible in every language and device context you intend to support.
Inspect the exported or stored data, not just the confirmation screen. Verify that the original rating, period, question version and appropriate context are retained. Check that a rating can be included without an optional comment and that a blank is not converted into zero.
Reproduce the expected score from the test data. Then make sure the test entries will not appear in the production report. Define who is responsible for investigating unexpected counts after launch.
8. Plan the review before sending invitations
Name the person or team that will read the results, review comments and decide what needs follow-up. Set a cadence suited to the survey’s purpose. A routine relationship survey does not automatically require immediate personal outreach to every respondent.
Where comments suggest a specific unresolved issue and contact is appropriate, connect the response to a responsible owner. For broader patterns, review other evidence before changing the service. Return an appropriate summary of what was learned and what the team will do.
Keep comments in context. They describe the respondents’ accounts, not proven causes of every rating. For a full action process, see using NPS feedback.
What makes a trend interpretable?
Retain the audience rules, dates, invitation channel, wording, valid counts and relevant group coverage for every wave. Show method changes beside the trend rather than burying them in a note no one sees.
An overall score can improve because more responses came from a higher-scoring group. Matching the same customers can help describe individual change, but does not by itself establish the cause or remove the effect of missing follow-ups. Report what was observed and where comparison is limited.
Use benchmarks only when their population and method are relevant. A target chosen without context can encourage score management rather than useful learning.
Choose a measurement workflow your team can maintain
Sopact’s relevant role is connecting collection, analysis and governance as feedback grows across sources and periods. Test the whole workflow with your team: verify a calculation, read a comment with its rating, review an interpretation and check what different users can access.
Keep the buying decision grounded in required work. A simple one-off survey may need little infrastructure. Recurring evidence across accounts, service records and teams may need stronger controls and clearer ownership. Compare implementation and maintenance effort as well as the subscription.
Watch the feedback companion
See Sopact’s discussion of reviewing feedback with its context.
Frequently asked questions
Is an open-ended follow-up required to calculate NPS?
No. The score uses valid recommendation ratings. An optional follow-up can help interpret the experience, but missing comments should not remove otherwise valid ratings.
Can I measure NPS anonymously?
Yes. Anonymous responses can support the score and group-level analysis. Individual follow-up and matching across waves require a separate, appropriate design.
How often should I send an NPS survey?
Choose a cadence suited to the customer relationship, change cycle and review capacity. Avoid repeatedly inviting the same people without a clear purpose.
Should passives be excluded?
No. They remain in the valid response total used to calculate promoter and detractor percentages.
Does consistent wording guarantee a valid trend?
No. Audience composition, timing, channels and response coverage also matter. Record those factors for each measurement.
