Skip to main content

​
Hub map

The Insights hub at /insights is a grid of roughly thirty-five tiles. This page is the map: one line per tile — the question it answers and when to open it — before you click, plus the task families the tiles group into, the roles that see each cluster, and the export paths out of every surface. The Insights hub overview walks the same grid top to bottom with worked examples; this page is the fast index.

​
What each tile answers

Each row names the tile, the route it opens, and the one question it exists to answer. For per-tile read guides, open the Section map on the hub overview. Cross-link tiles that live in sibling dashboards — Wallboard, Voice Quality, QA program, Performance Leaderboard — are listed there too; the badge reads “Lives in another section of the dashboard.”

​
The task families

Four families cover the grid. Pick a family before a tile — it narrows where to look.
  1. Comparison — Agent comparison, Attribution, Conversion Goals, Funnels & Retention. Decide which of two variants, channels, or steps wins. Start a rollout here.
  2. Cohort — Retention, Account Scores, Funnels & Retention. Confirm the comparison win survived its entry cohort. Route a variant through here before signing off.
  3. Operations — Logs, KPI Alerts, Usage anomalies, Usage alert rules, Scheduled Reports, Reports. Watch the live stream, arm the tripwires, and package numbers for stakeholders.
  4. Cost — Channel Costs, LLM spend, Sustainability, Agent ROI. Reconcile the budget per channel, model, and agent.
Cross-cutting reads — Conversation Intelligence, Contact Reasons, Topic Intelligence, Sentiment, Journey Paths, High-friction sessions — explain why a comparison or cohort read moved; open them in any family when the headline number needs a reason.

​
The decision loop

Operators run the grid as one loop:
  1. Compare two variants on Agent comparison.
  2. Confirm the win held per cohort on Retention.
  3. Assign credit on Funnels & Retention and Attribution.
  4. Pin the deciding metric with an alert on KPI Alerts.
The worked narrative — same four steps through a realistic exec decision — is in Read the Insights dashboards as one decision picture. The operator walkthrough with screenshots-is-reference numbers is on Build a comparison view across two variants.

​
RBAC — who sees which tiles

Tile visibility mirrors the normalised org role; the destination page re-checks it server-side, so the grid only hides what a role cannot open. Three clusters gate:
  • Finance cluster — Channel Costs, LLM spend, Sustainability, Agent ROI, Agent comparison: owner/admin/developer/billing seats.
  • CDP cluster — Retention, Funnels & Retention, Account Scores, Auto-drafted knowledge: owner/admin/developer.
  • Quality cluster (cross-link tiles) — Quality Management: owner/admin; Performance Leaderboard: owner/admin/supervisor; QA Evaluations: additionally agent, own evaluations only.
Everything else — Analytics, Dashboard Builder, Logs, KPI Alerts, Usage anomalies, Usage alert rules, Reports, Scheduled Reports, Campaign analytics, SMS Click-Through, Link Tracking, Deliverability, Conversion Goals, Attribution, Marketing mix model, Surveys, AI Containment, Per-language AI quality, Conversation Intelligence, Contact Reasons, Topic Intelligence, Sentiment, Journey Paths, High-friction sessions, Search Analytics — renders for every member, including viewers. Writes on alert rules stay with owner/admin/developer at the API; the read views are open. The full worked examples per seat are on the hub overview’s Role-specific views.

​
Export — three shapes, one per purpose

  • Ad-hoc snapshot (CSV/PDF). The Reports tile generates a report in the viewer toolbar — use it when a stakeholder asks for the numbers once.
  • Scheduled snapshot. The Scheduled Reports tile emails the same reports on a cadence to a recipient list — use it when the ask repeats weekly.
  • API / BI snapshot. Every tile’s surface resolves over the Insights API as JSON, so the same view lands in Tableau, Power BI, Looker, Superset, or Metabase through the CDP BI-tool connectors.

​
Troubleshooting

Two failure modes recur on the comparison and cohort families:
  • Attribution returns nothing, or everything lands in unattributed_conversions. A conversion with zero in-window touchpoints surfaces as unattributed, not dropped. If the share is large, sends are dropping the stitchable id (contact_id or anonymous_id) before the touchpoint lands — stamp the id at send time. Widen the lookback before changing the tracking plan: attribution joins per conversion across lookback_days (1–365, default 30).
  • Funnel and attribution totals diverge. The funnel counts events in its own window; attribution joins conversions against a per-conversion lookback — the scopes differ by design. When two stakeholders quote different numbers, align on model + lookback first, not on which tile is “right.”
Per-tile misreads — undefined CTR denominators, stale rollups, window-label mismatches — sit in each tile’s linked guide from the Section map.

​
See also