AI cost · Understand domain
AI token visibility, in one place.
Connect GitHub Copilot, OpenAI, Anthropic Claude, and Cursor in minutes. CloudMonitor blends them into a single cost view — model mix, cache hit rate, batch discount usage, and cost per developer — on the same ingestion path as your Azure spend.
The problem
AI tools bill in four places.
Spend scattered per vendor.
Copilot on one invoice, OpenAI on another, Claude and Cursor on two more. No single view shows what the AI tooling actually costs.
Seats bought, never used.
Paid Copilot and Cursor seats sit at single-digit acceptance rates. The waste is real, and invisible without a usage signal per seat.
Discounts left on the table.
Batch tiers halve cost; prompt caching cuts input cost roughly tenfold. Untracked, neither shows up in the bill you could have had.
How CloudMonitor answers
Four connectors, five minutes each.
Each connector backfills the last 14 days as proof before polling continues. Read-only by design: token counts, model names, and dollars — never prompt or completion text.
GitHub Copilot.
Business and Enterprise — per-seat utilization and acceptance, so dormant and low-acceptance seats surface for reclamation.
OpenAI.
Codex CLI, the API, GPT-5, the o-series, embeddings, audio, and vector stores — token counts and dollars only.
Anthropic Claude.
Claude Code, Sonnet, Opus, and Managed Agents — with prompt-cache and batch-tier usage broken out.
Cursor.
Business and Enterprise — spend and activity per developer through the Admin API.
AWS Bedrock, GCP Vertex AI, Azure OpenAI direct attribution, Windsurf, Aider, and JetBrains AI Assistant are on the roadmap — vote on the order.
On the dashboard
Five numbers, one AI cost page.
Total AI spend.
Dollar cost across every connected provider for the period, in one figure.
Cost per developer.
Total spend over distinct active developers, blended across every provider.
Model mix.
Share of spend by frontier, mid, and entry tier. A point moved from frontier to entry without quality loss is roughly 10× cheaper.
Cache hit rate.
Share of input tokens served from prompt cache — billed around a tenth of the standard input rate.
Batch percentage.
Share of spend run through batch APIs — 50% off on both OpenAI and Anthropic.
Behavioral surfaces (opt-in)
Turn the numbers into habits.
Opt-in per customer and anonymized by default — the insight without naming developers on a public board.
Efficiency leaderboards.
Ranked by output-tokens-per-dollar, not raw spend — plus prompt-cache and batch-tier champions per workspace. The top spot rewards the smartest user, not the heaviest.
Hidden-waste detectors.
Paid seats under 10% acceptance, seats with zero activity in 30 days, and Claude Code sessions left in Fast mode — the recoverable dollars most teams can't see.
Weekly digest & estimator.
A five-number summary lands in your Teams channel every Monday, beside a recomputed estimate of what shifting jobs to batch or a cheaper model would save.
Related Capabilities
More in the Understand Cloud Usage and Cost Domain.
Reporting and Analytics
Thirty-plus Power BI reports — the AI cost page lives alongside them.
Anomaly Management
Daily detection with alerts routed to Teams or any ITSM you already run.
All Understand Cloud Usage and Cost Capabilities
See every Capability in this Domain side by side.
See the AI cost page against a seeded tenant.
Connect a provider in five minutes, or walk the demo first. Included on every CloudMonitor plan — no add-on.
