Feedback to comparable insight

The classifier creates one Comments projection per free-text answer, labels sentiment and proposed themes, and stores masked display text. AI output is a suggestion, not a guaranteed classification or safety guarantee. CX Manager may correct permitted theme tags with provenance; recomputation counts canonical comments once. Anonymous rows contain no customer, ticket, agent or flow references. Raw text is not sent in notification bodies and private data is excluded from model requests unless the configured processor and policy explicitly permit it.

Responsible

CX Manager

Records

CommentsThemes

Effect

Create masked theme projections with classification provenance

Service recomputation uses Accepted responses and the frozen primary metric question once per respondent; skipped/invalid responses are excluded and zero denominators produce unavailable values. NPS category counts partition valid NPS respondents. CSAT favourable percentage uses its instrument range; CES keeps its explicit effort direction, applying the same polarity to easy/difficult shares, outreach and compatible benchmarks. Response rate divides distinct accepted completions by distinct provider-accepted invitations for the same matured occurrence population; queued, suppressed and failed sends are separate. Versioned cohort memberships and full PopulationKey prevent cross-campaign overwrite. Anonymous breakdowns respect minimum group size and forbid identity joins. Benchmark comparisons require matching metric, scale, population and period; incompatible or absent benchmarks remain unavailable.

Responsible

Analyst and CX Manager

Records

NPSScoresCSATScoresCESScoresCohortsCohortMembersTrendsBenchmarkData

Effect

Recompute versioned comparable score populations

Wording tests freeze both published template versions, stable invitation assignment, analysis method, split and minimum samples before Running. Variant denominators include assigned eligible sent invitations, not only respondents; retries do not move recipients. Reaching a sample threshold permits evaluation but does not itself prove a winner; report inconclusive/no significant difference under the prespecified method. Versioned results never replace ordinary campaign metrics. Alert only when the configured score-change threshold and minimum sample are both met; support-team alerts stay within authorized scope.

Responsible

Survey Author and Analyst

Records

SurveyABTestsSurveyInvitationsSurveyTemplates

Effect

Record the prespecified comparison and permitted alerts

Opt-out callbacks and authenticated requests upsert suppression immediately. Every queued send and reminder applies the current suppression, while completed evidence remains governed by retention and privacy requests. CX Manager closes/archives campaigns without editing accepted responses; audited invalidation triggers same-basis score recomputation. Admin preserves delivery keys and restricted correlation only for the approved retention period. Reports retain counts of invited, delivered, suppressed, accepted, invalidated and unresolved outcomes so unavailable data is never presented as a favourable score.

Responsible

CX Manager and Admin

Records

OptOutsSurveyResponsesTrends

Effect

Apply current consent and traceable recomputation