Skip to content
DMSurvey Codingby Decision Models Beta
Your workspaceDeutsch

NPS Verbatim Coding: Analyse Reasons for the Score

NPS verbatim coding structures the comments following a recommendation score. The number records a rating; the comment describes selected experiences. Keep both, and avoid turning a low score into a negative label for every theme in the text.

Keep the score separate from the explanation

Preserve respondent ID, original score, exact follow-up question and original comment. Create a separate set of theme columns for the text. If a respondent gives a score but leaves the comment blank, the score remains usable under your scoring rules. Missing text should not silently remove that respondent from the numeric NPS calculation.

The conventional grouping is promoters at 9–10, passives at 7–8 and detractors at 0–6; NPS subtracts the detractor percentage from the promoter percentage. The denominator includes valid scores in all three groups. The original scoring explanation describes this method. Theme coding does not change those score groups.

For the coding pass, consider hiding the score initially so it does not steer interpretation. Restore it for analysis after assigning themes. A score can help identify a case worth investigating, but it is not evidence that a comment contains an unstated topic or a particular sentiment toward each mentioned aspect.

For grouping, the original NPS scoring explanation uses promoters at 9–10, passives at 7–8 and detractors at 0–6. NPS subtracts the detractor percentage from the promoter percentage on the valid-score base. These groups organise the rating data; they do not replace coding the actual reasons.

Use themes that describe the experience

Start with categories tied to your service or product. Generic positive and negative buckets alone rarely tell an operational owner what to investigate. In this fictional subscription-service example, the follow-up question is “What is the main reason for your score?” Even though it asks for one main reason, respondents may give several, so the coding rules allow multiple themes.

Download the example NPS codebook CSV. It contains these six topical codes. Each is direction-neutral: positive and negative statements about the same aspect receive the same theme.

ID Label and meaning Include Exclude Example
value Price and value: money paid relative to benefit Price increases, affordability, worthwhile cost Time saved without reference to money “Worth the monthly fee.”
reliable Reliability: dependable operation Outages, failures, consistent operation Learning difficulty without a fault “It stops working every week.”
ease Ease of use: learning and completing tasks Setup, navigation, confusing workflows Missing capability rather than usability “New staff learn it quickly.”
service Customer service: support interactions Helpful agents, slow responses, unresolved tickets Product failure with no support contact “Support explained the fix.”
features Features and fit: available capabilities Integrations, missing exports, useful functions Difficulty finding an existing feature “I need an offline option.”
general General evaluation: unspecific overall judgement Good, poor, satisfied without a specific aspect Specific praise already captured by a theme “It is fine overall.”

Define the general and residual buckets carefully

Use general only when the answer gives an overall evaluation without a specific theme. “Great support” receives service, not general plus service. “Great support, but overall I am disappointed” receives both because the overall assessment adds a distinct statement. This rule prevents a generic code from appearing automatically on every positive or negative answer.

Use other for meaningful content outside the frame, such as “My employer requires this provider.” That is a reason for the score, but not an experience theme covered here. If such answers recur and matter to the research objective, add a defined category during the pilot and revisit earlier cases.

Use non_response for empty text, an explicit refusal or a placeholder that provides no reason. Do not use it for an unfavourable answer you cannot interpret immediately. In Survey Coding these two statuses are computed rather than uploaded as topical codes, so they are absent from the companion CSV.

Worked example: ten reasons for a score

These scores and comments are invented for instruction, not observations about customers.

ID Score Reason for score Theme codes
N01 10 “Dependable every day and easy for new colleagues.” reliable; ease
N02 4 “The fee rose again, and support never answered.” value; service
N03 8 “Does what I need, but the export to our CRM is missing.” features
N04 9 “Support was kind, although the app still crashes.” service; reliable
N05 6 “Too many screens to finish a simple task.” ease
N06 10 “Worth what we pay each month.” value
N07 7 “Fine overall.” general
N08 3 “My employer requires this provider.” other
N09 8 “I would rather not explain.” non_response
N10 5 “The offline mode is useful, but outages keep interrupting me.” features; reliable

N04 is particularly useful in a pilot: a high score accompanies both helpful service and a reliability complaint. N10 combines a valued capability with a technical problem. The theme assignment preserves both aspects. If you add direction, service in N04 is positive and reliability is negative; neither should inherit the overall score's group label.

N03 mentions fit and a missing integration. The wording does not demonstrate that navigation is difficult, so ease is not added. N08 remains substantive even though its reason is outside the frame. N09 keeps its numeric score while contributing no usable reason to the text analysis.

Review before comparing score groups

Pilot the frame on comments from all score groups, including mixed and short answers. Review the definitions without assuming that promoters only praise and detractors only complain. Flag sarcasm, comparative statements, unclear references and comments that appear to address a different experience from the follow-up question.

If the codebook changes, recode affected earlier comments. Keep a record of version, reviewer decisions and any direction rules. Inspect a random sample outside the uncertain queue as well: an apparently obvious phrase can still be assigned to the wrong aspect. Retain rare, serious complaints for investigation even if they do not justify a separate reporting category.

Make tables that answer a limited question

Cross-tabulate themes by score group and show counts beside percentages. Label whether the base is all scored respondents, respondents providing substantive comments or respondents with an assigned topical code. The share of people mentioning service among commenting detractors is different from the share among all detractors, including those who left no text.

Compare comment availability across groups before interpreting theme differences. Voluntary explanations may represent different subsets of promoters and detractors. Avoid calling a frequently mentioned theme a proven driver of NPS: an association between a comment and a score does not establish what caused the score or what an intervention will change.

Give operational owners code definitions, anonymised examples and the relevant base. Treat frequent themes as questions to investigate, without promising a score increase.

Short checklist

  • Retain all valid scores independently of comment completeness.
  • Apply themes from the words, without inheriting score sentiment.
  • Permit multiple themes and define general versus other.
  • Review examples from every score group.
  • Label comment bases and keep counts beside percentages.
  • Present associations as investigation leads rather than causal proof.

FAQ

Should I code only detractor comments?

Include all groups when the objective is to understand reasons across the customer base. A detractor-only project is possible, but label that restricted scope and avoid generalising its theme shares to everyone.

Is a promoter comment always positive?

No. A promoter may mention a specific complaint, and a detractor may praise one aspect. Code the text and retain the score separately.

Does a missing comment change NPS?

A missing explanation does not itself invalidate an otherwise valid score. Use separate rules and denominators for the numeric result and the text analysis.

Should a rare complaint get its own code?

Consider the research objective and whether the category can be applied meaningfully. You can preserve and investigate a serious individual case without creating a tiny reporting category.

Survey Coding by Decision Models (Beta) accepts CSV/XLSX with a codebook or template, provides probabilities per code and supports review of uncertain answers. Results export as CSV/XLSX and CSV with SPSS syntax; up to 500 answers per project are free. Try Survey Coding.

Codebook CSV ↓

GuidesDecision Models Survey Coding · Beta
Data handlingImprintPrivacyTermsSurvey Coding Additional TermsDPA Annex S