Skip to main content
Vorel ships a per-tenant analytics dashboard at /(dashboard)/analytics and a programmatic weekly-rollup API for around-the-brain digest workflows. Same computeAnalytics() helper backs both; the route runs under withTenantContext so RLS scopes every read.

​
What the dashboard surfaces

app.vorel.ai/(dashboard)/analytics: five cards over the last 30 days:

Conversation volume

Total conversations + breakdown by channel (whatsapp / voice) + breakdown by status (open / closed / handoff_pending / etc.). Server-rendered; loads in a Suspense boundary against the same RLS-scoped tx.

Lead funnel

Conversations started → leads created → leads moved off new → appointments booked. Four counters; raw conversion math is up to you.

QA score distribution

Average score over the window + a 4-bucket histogram (0-3 / 4-6 / 7-9 / 10-11 on the 11-criterion rubric). Empty buckets render as zeros so the histogram axis is stable.

Tool-call success rate

Top 20 tools by call count, with success rate computed from messages.tool_payload->>'ok'. Fast surface for “is crm_create_record flapping for this tenant this week?”

Voice-minute trend

Per-day voice-minute consumption, oldest-first, with empty-day zeros so the chart doesn’t gap. Total minutes summary on top.
The dashboard requires the user’s authenticated dashboard session to resolve a tenant. Cross-tenant operator views (/admin/tenants/[id]/analytics) reuse the same computeAnalytics() function with the operator’s tenant pick.

​
The weekly rollup (digest workflows)

GET /api/v1/analytics/weekly-rollup: JSON body shaped for the weekly-qa-rollup automation template (Monday-09:00 cron, then format, then deliver by email or Slack). Response shape: 
{
  "since": "2026-04-29T00:00:00.000Z",
  "computed_at": "2026-05-06T09:00:00.123Z",
  "qa_avg": 7.4,
  "qa_p50": 8.0,
  "qa_p95": 11.0,
  "conversation_count": 142,
  "agent_turn_count": 1031,
  "hallucination_flag_count": 7,
  "top_flag_kinds": [
    { "kind": "unsupported_price", "count": 3 },
    { "kind": "unsupported_availability", "count": 2 }
  ],
  "top_handoff_reasons": [
    { "kind": "explicit_request", "count": 12 },
    { "kind": "complaint", "count": 4 }
  ]
}
Window contract: every metric in the response is created_at >= since. Default since is 7 days before computed_at. Maximum window is 90 days; wider windows return 400 bad_request to avoid sequential scans across the full messages table.

​
Authentication + scopes

Public-API analytics endpoints are bearer-token authed:
  • GET /v1/analytics/weekly-rollup: read scope. Per-key rate limit 200 req/min.
Authed via vapk_live_<48 hex> API keys issued from /(dashboard)/settings/integrations/api-keys. The dashboard surface (/(dashboard)/analytics) uses your authenticated dashboard session; no API key required for the in-app view.

​
Data sources

  • conversations: channel + status counts.
  • leads: funnel counters (status + created_at for the window).
  • appointments: appointments booked in the window.
  • qa_evaluations: score + per-criterion breakdown for QA.
  • messages: tool-call success rate via tool_payload->>'ok'; agent-turn count for the rollup; hallucination flags for the digest.
  • billing_events: voice-minute trend (sum of voice_minute event types per day).
All reads use the RLS-scoped tx, so a tenant only ever sees their own data.

​
The analytics workbench

Beyond the at-a-glance dashboard cards, Vorel ships an analytics workbench: saveable reports built from a report definition (source, metric, dimensions, filters, visualization, time window), composable dashboards, scheduled CSV / email-summary exports, threshold alerts, read-only share links, and built-in preset rollups. A natural-language composer (the in-product AI rail) turns a plain-English ask into a validated report definition. Power users can also drop into a guarded read-only SQL escape hatch. Every report compiles against a per-source field allow-list and runs under row-level security, so a report can never read past your tenant scope. Custom date ranges, compare-to-previous-period, and CSV export are supported through the workbench and the scheduled-export path.

​
What’s NOT supported today

  • Per-conversation drill-down from a chart. Click-through from a histogram bucket to the underlying conversations isn’t wired yet.
  • PDF export. CSV and email-summary exports ship; PDF is not a current export format.
  • Real-time streaming updates. Surfaces are server-rendered on each request, not live-streaming.

​
Around-the-brain integration patterns

The weekly-qa-rollup template (n8n templates) wires GET /v1/analytics/weekly-rollup → format → email/Slack on a Monday-09:00 cron. Operator imports the template into n8n, sets the API key + delivery target, and the digest fires automatically. For custom analytics workflows (per-segment cohorts, per-month trends, per-vertical comparisons), build against the public API + your own warehouse. Vorel exposes the canonical numbers; your analytics stack does the rest.