Governance usage & cost dashboard
Monitor AI capability spend, token usage, CO₂ estimates, and chargeback reports from the Governance settings page.
Organisation admins can view a live cost and usage breakdown for every AI capability call made by their organisation — broken down by workspace, team, user, feature, vendor, model, or region — directly from Settings → Governance → Cost Insights.
The whole Governance section (General, DLP, Clearance, and Cost Insights) is visible only to organisation admins. Members who navigate to /settings/governance/cost-insights are redirected to their account settings.
Opening the dashboard
- Sign in as an organisation admin.
- Open Settings and select Cost Insights under the Governance section in the sidebar.
The dashboard loads with a default 30-day window, grouped by team.
The dashboard loads with a default 30-day window, grouped by team.
Filter bar
At the top of the dashboard you will find two controls that apply to all five views:
| Control | Options | Default |
|---|---|---|
| Date range | Pick any calendar range (single or multi-month) | Last 30 days |
| Group by | Team · Workspace · User · Feature | Team |
Changing either control refetches all data server-side immediately. A subtle "Updating…" indicator appears while the fetch is in progress.
Views
Overview
The landing view shows four hero stat cards:
- Total spend — cumulative cost (USD) for the selected range, with run count.
- Executions — total workflow runs and success rate.
- Tokens — total tokens consumed (millions), split into input and output.
- Carbon footprint — modelled CO₂ equivalent in kg, with driving-distance and tree-year analogues.
Below the stats, a Spend over time area chart plots daily cost. Two breakdown tables show Cost by model (top 6) and Top workflows by cost (top 5).
Spend figures are indicative estimates, not invoices. Token counts are the
exact values reported by each model provider. Cost is then derived in priority
order: (1) the exact pricing configured on the integration/model, (2) the base
model's pricing when a dated deployment is used (e.g. an Azure deployment
gpt-5.4-2026-03-05 reuses the gpt-5.4 rate), then (3) a configurable
default rate when a model has no catalog pricing. Cost is always labelled
indicative and is intended for internal governance and chargeback, not billing
reconciliation.
Spend
A tabular breakdown of cost, tokens, and runs for every group in the selected dimension (team / workspace / user / feature). Columns are sortable.
Execution
Run-level metrics: total runs, success rate, and failure count. A sortable table lists the top workflows by cost for the period.
Carbon
CO₂ figures are modelled estimates — energy-per-token × datacenter PUE × regional grid carbon intensity. They are suitable for internal governance and sustainability reporting, not for certified environmental disclosures.
The Carbon view shows:
- Estimated carbon footprint headline in kg CO₂e.
- Equivalents: kilometres driven and tree-years needed to absorb that amount.
- Methodology disclosure — the carbon-factor table version and citation. The factor table is updated periodically; the version displayed matches the table that was active when the data was recorded.
Chargeback
A cost-allocation table suitable for internal chargebacks, showing each group's absolute cost, percentage share of the organisation total, and CO₂ contribution.
The Export CSV button downloads the current table as a comma-separated file (chargeback-<dimension>.csv) with columns: department, cost_usd, share_pct, co2_kg, runs.
The export button is disabled while the table is loading or empty.
Configuration
The Configuration view is where org admins set the indicative fallback pricing used when an integration ships no pricing for a model. It has two sections:
- Model price overrides — pick an installed model from a dropdown (grouped by capability: LLM, Embeddings, Speech-to-Text, Text-to-Speech) and set its price. Token-priced models (LLM, embeddings) take a
$ / 1M tokensinput rate (LLM also takes an output rate); unit-priced models take a$ / unitrate. A dated deployment such asgpt-5.4-2026-03-05automatically reuses the basegpt-5.4rate. - Capability defaults — a fallback rate per capability (LLM, Embeddings, STT, TTS, OCR, Image, Video), used when neither the integration catalog nor a model override prices a call. Leave a field blank to keep the built-in indicative default (shown as a placeholder). This is how capabilities without an enumerable model list — OCR, image, and video — are priced.
An integration that ships its own pricing always wins. Configuration rates are only a fallback. Resolution order is: (1) exact catalog/integration pricing, (2) the base-model catalog rate for a dated deployment, (3) a per-model override from this tab, (4) the per-capability default (configured here, else built-in). Cost stays indicative and applies to future usage — already-recorded rows keep the price that was frozen when they were logged.
Permissions
| Role | Access |
|---|---|
| Organisation owner | Full read access to all views and CSV export |
| Organisation admin | Full read access to all views and CSV export |
| Member | Redirected to account settings |
The underlying client.billing.* API requires an org-admin session. See Billing and usage metering for the full API reference.
Data freshness
Usage events are written by the capability runtime at call time. The dashboard queries are live — there is no scheduled aggregation delay. Events from capability calls that occurred within the selected date range are included immediately.
A failed billing write never blocks the capability call itself. If a brief write outage occurred, a small gap in the timeseries is expected and does not indicate a product failure.