Skip to main content
The Executive Dashboard is the top-of-stack view for leaders who want the story, not the raw numbers. It arrives with a weekly narrated summary, six period-over-period KPI cards, an anomaly feed, and a segment × offer heatmap — everything you need to spot what changed this week in 30 seconds. Path: /dashboards/executive

​
What’s on the page

​
1. Weekly narrated summary

A short 1–2 paragraph LLM-generated executive summary appears at the top of the page on every load. The narrative is produced by the report runner using the same data sources the page surfaces; it runs server-side, is cached for one hour per tenant, and regenerates automatically on the next page load after expiry. If the LLM provider is unavailable or your tenant has no AI provider configured, the card displays a fallback message (“Narrative is unavailable right now — see the KPI cards below for the raw numbers”) and the rest of the page continues to work.

​
2. Six KPI cards (3 × 2 grid)

Each card uses the PeriodDelta component: a headline value, a colour-coded Δ% vs. the prior period, and (where applicable) a tiny sparkline.
CardSourceWindowFormat
Active offersGET /api/v1/dashboard-data?type=summary_with_comparisonPoint-in-timeInteger
Decisions (7d)Same — decisions.{current,previous} + sparklines.decisions7dInteger
Acceptance rateGET /api/v1/dashboard-data?type=acceptance_rate&days=7 vs. days=147d vs 7dPercent
Revenue (7d)summary_with_comparison.revenue + sparklines.revenue7dCurrency
Model AUCGET /api/v1/dashboard-data?type=model_auc_summary_with_prevLatest vs prior evalFloat
Anomalies (7d)GET /api/v1/dashboard-data?type=anomaly_candidates&days=77dCount
Delta colour conventions:
  • Green (emerald) — positive change vs. prior period
  • Red — negative change
  • Grey flat chip — within 0.05% of prior value
  • Blue “New” chip — prior period was zero (no baseline to compare)

​
3. Anomaly feed

The left panel shows the last seven days of anomalies detected across core metrics (acceptance rate, latency, degraded scoring, selection frequency). Each row carries:
  • Severity pill (info / warning / critical)
  • Human-readable metric + dimension
  • Δ% vs. baseline and z-score
  • Explain button → opens an inline popover with a narrative explanation of the anomaly. Calls POST /api/v1/ai/explain when wired; today the route is not yet exposed, so the button falls back to a deterministic summary built from the row’s fields. Wiring the full LLM narrative path is tracked in the roadmap.
Empty state: “No anomalies in the last 7 days. All metrics within expected ranges.”
The anomaly feed reads directly from the anomaly_candidates endpoint on every page load — it does not depend on the alert-rule cron being wired. Anomaly rows appear here whether or not you have alert rules configured and whether or not an external scheduler is calling /api/cron/tick.

​
4. Segment × Offer heatmap

The right panel visualises selection rate across segments × offers. Rows = top 10 segments by customer count; columns = top 10 offers by scored count. Cell colour keys off the selection rate (emerald gradient); hovering surfaces the raw scored / selected counts. If a tenant has no segments configured, the heatmap shows a stub with a deep link to /data/segments rather than rendering an empty grid. At the bottom of the page, four deep links to the corresponding operational dashboards (Business / Operations / Model-Health / Attribution) let you jump from the narrated summary straight into the detail view.

​
Sharing

The top-right header of the Executive Dashboard (and every other dashboard) has two buttons:
  • Export — dropdown to download the current view as PDF / CSV / Markdown / HTML. Under the hood this calls the transient POST /api/v1/reports/preview endpoint which runs the report runner without persisting a template.
  • Save as Report — opens a modal pre-filled with the current view’s data sources and filters. Pick formats, narrative on/off, schedule (daily 8am / weekly Mon 8am / monthly 1st 8am / custom cron), and destinations (any configured notification provider). Saving creates a ReportTemplate and a ReportSchedule.
Export works unconditionally and requires no cron wiring. Clicking Export renders the artifact server-side and streams the file back to the browser immediately. You can use it at any time — no EventBridge, no CRON_TOKEN, no background automation needed.Save as Report also creates the template + schedule unconditionally, but the resulting schedule only fires on cadence once the cron is wired. Until then, use the Run Now button on the saved template in /settings/reports for on-demand delivery. See EventBridge Setup for the optional automation path.
See the Share a Dashboard guide for an end-to-end walkthrough.

​
Performance

  • KPI cards: load in < 500 ms (react-query parallel fetch of 4 endpoints).
  • Narrative: 1–3 s on first load; cached for 1 hour per tenant.
  • Anomaly feed and heatmap fetch in parallel with the summary and populate as they arrive (skeleton states prevent layout shift).

​
Tenant isolation

Every backing endpoint calls requireTenant() before issuing queries. Anomaly rows, segment lookups, selection frequency aggregations, and the narrator’s data-source invocations all filter by tenantId — no cross-tenant data leaks on shared deploys.

Dashboards

Operational dashboards powering each of the quick-links on this page.

Reports

Templates, schedules, formats, and delivery — the engine behind Save-as-Report.

AI Insights

Cross-decision anomaly helpers and the narrator that powers the weekly summary.