What is the difference between NPS and CSAT?
NPS asks how likely someone is to recommend a company, product or service. CSAT asks how satisfied they are with an experience or offering. NPS uses a standard 0–10 recommendation scale and a promoter-minus-detractor calculation. CSAT commonly reports the percentage selecting satisfied responses on a satisfaction scale.
Choose the measure that fits the decision. CSAT is often useful after a particular interaction; relationship NPS can provide a broader view of willingness to recommend. These are common uses, not rigid boundaries: NPS can also be collected around a transaction, and CSAT can ask about overall satisfaction.
Neither score explains everything. Read the accompanying comments and relevant operational evidence to understand what respondents experienced. A low score is a reason to investigate, not proof of a specific cause or a certain cancellation.
NPS vs. CSAT at a glance
| Question | NPS | CSAT |
|---|---|---|
| What does it ask? | Likelihood of recommending an organization, product or service | Satisfaction with the specified experience or offering |
| Typical scale | 0–10 | Often a five-point satisfaction scale; other formats exist |
| Common calculation | Percentage scoring 9–10 minus percentage scoring 0–6 | Percentage selecting the responses defined as satisfied |
| Range under that calculation | −100 to +100 | 0% to 100% |
| Useful application | Reviewing recommendation intent across relationships or relevant experiences | Examining satisfaction with a service, product or interaction |
| Important limitation | Intent is not actual referral or retention behavior | Satisfaction with one experience need not describe the entire relationship |
Keep the wording and context visible in every report. “How satisfied are you with the support you received today?” and “How satisfied are you with our product overall?” are both satisfaction questions, but their answers should not be treated as the same measure.
How to calculate NPS
Classify valid 0–10 ratings as promoters (9–10), passives (7–8) and detractors (0–6). Subtract the detractor percentage from the promoter percentage. Passives remain in the denominator even though they are not in either term of the subtraction. See Bain's NPS scoring definition.
Illustrative calculation: among 200 valid ratings, 100 are promoters, 60 are passives and 40 are detractors. NPS = 100/200 × 100 − 40/200 × 100 = 50 − 20 = +30. Report it as an NPS value, alongside the count and distribution.
Do not calculate NPS by averaging the ratings or by dropping passives. An average recommendation rating can be reported separately, but it is not the same statistic. Exclude missing ratings from the valid-rating calculation and disclose missingness separately.
How to calculate CSAT
Define which response options count as satisfied before analyzing the data. On a five-point scale, the top two categories are commonly used: satisfied and very satisfied. Divide their count by the valid responses to that item and multiply by 100. This is the common top-two-box CSAT calculation.
Illustrative calculation: 160 of 200 valid responses select satisfied or very satisfied. CSAT = 160/200 × 100 = 80%. Retain the full distribution so a change from very satisfied to satisfied is not concealed by an unchanged headline percentage.
Document any different scale or scoring rule. Do not compare a top-two-box five-point score with a yes/no satisfaction score as though the measurement were identical. Keep not-applicable and missing answers distinct, with their treatment recorded.
Which measure should you use?
Start with the action the team needs to take. If a support manager wants to understand the experience of a recent interaction, a clearly scoped CSAT question may fit. If leadership wants a recurring view of recommendation intent, a consistently designed NPS program may fit.
A program or service team should also ask whether recommendation is a meaningful question for its audience. People may have little choice of service, or may not be in a position to recommend it. Direct questions about access, usefulness or a specific experience can sometimes be more informative.
Use both measures only when each supports a distinct decision and the organization can use the extra information. More measures are not automatically a better listening program. Coordinate requests so one person does not receive several surveys from different teams after the same event.
When should you ask?
Ask after respondents have enough experience to answer, and while they can recall the relevant details. A request immediately after a support interaction can assess that interaction, but it may arrive before the person knows whether the issue stayed resolved. Decide which experience you mean to measure.
For a recurring relationship survey, use a defined population and a consistent approach to timing. Avoid surveying only people who recently had a successful interaction if the goal is to understand the overall relationship. Keep exclusions and invitation rules visible.
Record changes in channels, timing or question wording. A new response pattern after switching from email to an in-app survey may reflect the method or audience, not only a change in experience. The collection plan belongs with the results.
How to use NPS and CSAT together
Keep each response's identifier, question, rating, date, reference period and relevant context. Where appropriate and permitted, connect responses to the same customer or account. Retain the distinction between an interaction-level response and a relationship-level response.
A fictional customer might report high satisfaction with a helpful support agent but low recommendation intent because the product still does not meet their needs. Another might recommend the product while reporting a frustrating billing interaction. Those combinations are informative rather than contradictory errors.
Do not average NPS and CSAT into a single score. Compare the patterns and supporting evidence. If you relate a particular interaction to a later relationship rating, define the time window and account for repeated interactions, missing responses and differences in who answered.
Anonymous feedback may support group comparisons without a customer-level link. Explain that limit instead of trying to infer identities. A linked design needs clear expectations about access and use.
Use comments to understand what the scores miss
An optional open question can reveal which experiences respondents considered. Code the answers using defined themes, preserve the supporting passages and permit multiple themes when one comment describes several issues. Keep the original wording available for review.
Compare themes within relevant score groups and customer segments. Do not assume a promoter has no complaints or a detractor has no positive experiences. The comment may describe a narrower event than the question intended; that itself can help improve the survey design.
Report how many rating respondents also supplied usable comments. If 200 people give ratings and 90 leave comments, the theme analysis does not represent all 200 equally. State the comment denominator, and avoid presenting a selected quotation as evidence of a universal experience.
When the codebook changes, version it and review whether earlier comments need reprocessing. Otherwise, a new theme definition can look like a new customer problem. The detailed workflow is covered in NPS verbatim analysis.
Turn the comparison into a sensible priority
Combine the frequency of an issue with its severity, context and the team's ability to address it. A frequent minor inconvenience and a less common but serious barrier should not be ranked by mention count alone.
Check operational evidence where it is available and appropriate: unresolved service issues, repeat contacts, setup completion or later cancellations. Define the relationship you are testing. A correlation between a low score and cancellation does not establish which experience caused the customer to leave.
Assign an owner, next action and review date. Compare later evidence with the earlier period using compatible definitions, while checking for changes in respondent mix. If the evidence remains unclear, state that and decide what further information would help.
Common comparison mistakes
- Treating 80% CSAT and +30 NPS as points on one scale. They use different questions and calculations.
- Comparing unlike populations. Support users are not necessarily representative of all customers.
- Ignoring small groups and missing responses. Show the denominator and uncertainty before interpreting a change.
- Calling a score a churn prediction. Test it against actual behavior before making that claim.
- Replacing the metric with sentiment. Text sentiment and a stated rating can differ for valid reasons.
- Collecting two measures with no owners. Each measure needs a decision and a review process.
Where a connected workflow saves effort
Sopact brings collection, record context, text coding and quantitative analysis together. The team can retain the relevant customer history, keep comments with their ratings and review source evidence behind a reported result. That helps when repeated surveys otherwise create separate files that someone must reconcile for each new question.
The team still defines the measurement and the codebook. Configured analysis can apply those definitions across eligible records and reprocess affected comments after a revision, while people check exceptions and interpretation. The useful comparison is the repeated work removed, not a claim that other platforms cannot analyze text.
Measure setup, coding, revisions, joins, review and reporting effort across a realistic year.
Frequently asked questions
Is NPS better than CSAT?
Neither is universally better. Use recommendation intent when that is the question you need to answer, and satisfaction when you need to understand the specified experience. Keep the scope, population and planned action clear.
Can NPS be transactional?
Yes. An NPS question can be asked in connection with a transaction or experience. Document the context and do not treat that result as directly interchangeable with a broader relationship survey.
What is a good score?
Interpret the result against your own consistent history and genuinely comparable benchmarks. Industry, audience, geography, question wording and collection method can affect comparisons. A headline threshold cannot replace that context.
Should we read every comment?
Aim to analyze the eligible evidence set rather than quietly selecting only convenient comments. Automation can help apply categories, with a planned human review of quality, exceptions and uncommon themes. Disclose any sampling or exclusions.
Can we link the results across tools?
Yes, if the identifiers, dates, permissions and source fields support a reliable connection. Test matching and corrections, retain unmatched records and measure the ongoing maintenance effort. For options, see feedback analytics software.
Choose your next step
King Center
Read a customer example of learning from collected feedback. This is not an NPS performance benchmark.
Read the customer story →