Skip to main content

Investigate Friction Insights

Friction Insights groups repeated evidence into themes. Grouping, confidence, weighting, impact, and recommended actions are heuristics—not statistical proof or verified root causes.

Investigation workflow​

Select period
→ apply context filters
→ inspect theme
→ inspect representative evidence
→ open original Captured Question
→ decide likely action
→ revisit a later period

Server-enforced scope​

The Insights request sends exact filters to the server for:

  • context surface;
  • embed host;
  • page path;
  • UTM campaign; and
  • widget version.

The server validates and applies these to the inspected conversation records, not only to visible cards. Date range, comparison, confidence, clustering, evidence depth, and cross-assistant aggregation are also plan-controlled.

Read the evidence correctly​

  • Surface rollups describe validated context values.
  • Failure reasons aggregate reporter-selected downvote reasons.
  • Representative questions/snippets are examples, not a complete census.
  • Follow-up/unanswered state is derived from conversation evidence.
  • Period comparison describes a selected window versus the preceding window; it does not show that a product change caused the difference.

Open at least one original record and inspect question, response, context snapshot/journey, and feedback before creating work. Use webhook handoff only after validating the theme; see Webhook Handoff.

If filtering looks surprising, clear one filter at a time, confirm exact host and path values, and use Troubleshooting.