Understand the dashboard

Steps

1. 1MTD spend tile: total cost across every connected provider, attributed by user/team/project. Click through to /finance/users for the per-user breakdown.
2. 230-day trend chart: stacked area chart of daily spend, segmented by team or model. Forecast extends to end of month using the last 7 days' run-rate.
3. 3Top spenders panel: who's burning the most, sorted by current month. Click a row to drill into their per-model breakdown.
4. 4Anomalies queue: things our anomaly detector flagged as 'this is 3x your normal pattern'. Most are real (a misconfigured loop, a launched new agent) — click in to confirm.
5. 5Approvals pending: count of requests waiting on an admin decision. Click through to /finance/requests.
6. 6AI subscriptions panel: every recurring AI charge we know about — ChatGPT Plus, Cursor, Claude Pro, etc. Click 'Full inventory' to see the per-vendor breakdown.

Watch-outs

• Forecast is run-rate-based, not seasonally-adjusted. If your AI spend has a weekly pattern (e.g. heavy weekdays), the forecast may overshoot.
• Anomalies are sensitive to small absolute numbers — a service with $10/day average can fire false anomalies on a $30 day. Mark these as 'expected' to teach the detector.
• Dashboard data is denormalised for speed. If you see a number that disagrees with /finance/users, the per-user view is the source of truth — refresh the dashboard and the rollup will catch up within 5 minutes.