Skip to content

Measurement honesty

UserVane sells a number you can put in front of leadership. That only works if the product refuses to look more certain than the sample allows.

For NPS and related headline scores:

  • Below 10 responses, no headline number. You still see responses and an explicit “not enough yet for a stable score” state.
  • In the still-stabilizing band above that, the score is shown with a sample margin (deterministic, not model-generated). Binary hide/reveal alone is not enough: at low N one respondent can move NPS by a large amount.

A triggered, targeted, sampled survey is not a pure random sample. The read surfaces:

  • The segment / targeting context
  • The denominator (N)
  • Response-rate context where available

Never a naked “your NPS” stripped of who was asked.

People who saw an ask and did not answer are not the same as people who never saw one. Honest product copy treats non-response as a named limitation of what the score covers, not as invisible missing data that you can ignore while claiming representativeness.

Responses store the survey version answered. Scoring is per version. Rewording a question bumps the version so you never silently pool incompatible wording into one lifetime number.

Absolute response caps and quiet periods are decided when the show-token is issued. If someone already saw the survey and holds a valid unused token, submit accepts the answer. Enforcing a cap at submit would throw away real feedback after the interruption: the worst outcome for a respondent-respect product.

Answered and dismissed state is server-authoritative, keyed by respondent id, returned on bootstrap. That is what makes “never re-prompt an answerer for the life of that survey” true across devices. Client local storage is a backup only.

After a DSAR erase, a hashed suppression tuple is retained so never-re-prompt survives deletion of the verbatim. See Privacy.

Free-text is stored and shown as written. It is end-user-authored PII by default. UserVane does not LLM-summarize verbatims into claims the respondent did not make.