How to analyze NPS comments without treating sentiment as the whole answer
Analyze NPS by calculating the score from valid responses and reporting counts, response rate, and distribution, then code comments with a reviewed multi-label taxonomy. Compare themes across promoter, passive, and detractor groups and relevant customer context, while keeping nonresponse and selection bias visible.
Clues in the source data
- The NPS score appears without response count or rate.
- Sentiment is treated as the same thing as promoter/detractor group.
- Comments are summarized without customer lifecycle or product context.
Why this gets messy
A sentiment model can label a polite complaint positive or a constructive promoter comment negative. NPS respondents are not always representative, and account size can make a small number of responses commercially important but statistically unstable.
From the field · Customer-experience manager
Explain an NPS drop after a product migration
NPS falls after a migration, but response rate also changes. Comments mention onboarding, missing features, speed, support, and pricing. Enterprise accounts are few but commercially material.
Columns used in the example
response_id · account_id · score · comment · survey_date · invited_flag · plan · migration_cohort · account_value
Follow it with your own file
- Reconcile invitations, valid responses, duplicates, and response rate by cohort.
- Calculate NPS components rather than only the net score.
- Code migration, feature, performance, support, price, and praise themes.
- Compare themes and operational evidence by cohort and plan, with counts and privacy controls.
Check: Promoter/passive/detractor counts must reconcile to valid scores, the NPS formula must reproduce, and conclusions must disclose response-rate and sample changes.
A method you can follow
- Validate the survey: Check eligible population, invitations, valid score range, response timing, duplicates, and nonresponse.
- Calculate transparent score evidence: Publish promoter, passive, detractor counts, percentages, and response rate.
- Code themes: Allow several themes and distinguish problem, praise, request, effort, and resolution context.
- Prioritize carefully: Combine prevalence, account impact, recurrence, and supporting operational evidence.
Choose the approach that matches the evidence and repetition
| Approach | When it works | What to verify |
|---|---|---|
| NPS score | Compact relationship signal | Diagnosis and nonresponse |
| Sentiment | Comment triage | Nuance and survey score meaning |
| Score plus themes/context | Experience prioritization | Taxonomy and sampling limits |
NPS is a compact signal, not a diagnosis
The score shows the balance of response groups under one survey design. Comments and operational evidence explain what respondents experienced, while nonresponse and customer mix bound what can be generalized.
Where this usually goes wrong
Do not weight or publish individual account comments without a documented purpose, access control, and privacy review. Commercial importance does not remove confidentiality obligations.
The short version
- Check eligible population, invitations, valid score range, response timing, duplicates, and nonresponse.
- Publish promoter, passive, detractor counts, percentages, and response rate.
- Allow several themes and distinguish problem, praise, request, effort, and resolution context.
- Combine prevalence, account impact, recurrence, and supporting operational evidence.
Doing these steps in Prepyr
Upload a CSV working copy in Prepyr. Describe the result you need: Analyze NPS scores and comments with themes response rates and context. Check the columns selected, preview any proposed changes, and compare the output with the source before applying or exporting it.
Prepyr steps used here: Text analysis · Charts · Reports · Sensitive Column protection
Final checks before using the result
- Eligible population checked
- Response rate shown
- NPS components reconciled
- Themes multi-labeled
- Sampling and privacy limits stated
An NPS report that explains the score movement with transparent response evidence and actionable themes without reducing every comment to sentiment.
Questions readers usually ask next
Is NPS the same as sentiment?
No. NPS is calculated from a recommendation score; sentiment is an interpretation of text. They can disagree and should remain separate.
Why include response rate with NPS?
A score from a changing or small respondent population may not represent the invited customer base consistently.