AI FinOps
Copilot, Claude or ChatGPT seats, Azure OpenAI, Bedrock or Vertex tokens, reserved capacity: we measure what your AI costs, attribute it to teams and uses, and apply the FinOps Foundation's token economics framework.
This is your situation if
- Your AI invoices rise faster than the uses you know of
- Seats bought, without knowing how many are active
- An AI project has to go through the committee with no estimated production cost
- Nobody owns the line: neither IT, nor finance, nor the business
Format
Status: offer in pilot; the format and the duration are set after the first pilots.
Embedded: monthly tracking, inside your teams.
Fixed scope: a baseline in the short format of the audit, then the instrumentation.
Your key stays with you: when a cost API requires an admin key, extraction runs on your side and the key is never passed to us.
The cost of AI, by team and by use
A baseline
By consumption mode, by team and by application.
Seats reallocated
Dormant seats listed, reallocated before any repurchase.
Cost per model and per channel
Cost per million tokens compared, discounts and commitments included.
Reserved capacity or pay as you go
A quantified recommendation on your real utilization rate.
Usage rules
Model suited to the task, cache, caps and alerts.
Showback by team
Cost per active user and per thousand tokens.
The same framework, three uses
Internal allocation, governed uses
AI costs by entity and by team, budgets and alerts, agents placed under an owner.
Large enterprisesCost per client and per feature
For software vendors and scale-ups that sell AI: margin per client, choice of models.
SMEs and scale-upsRebill AI to your clients
Showback by client from the start of tracking; rebilling with margin will come with the AI costs module, planned.
The planned moduleAlready practiced in the field
- Claude Enterprise and Microsoft 365 Copilot seats: real activation against seats bought, dormant seats, reallocation rather than removal
- Azure OpenAI tokens by deployment, model and application; trade-off between reserved capacity (PTU) and pay-as-you-go
- Cost per million tokens compared by model and by access channel: AI spend put under governance rather than cut, within a multi-cloud execution task force (client case)
Frequently asked questions
Why an offer "in pilot"?
The frame and the levers are practiced in the field, but the duration and format will be set after the first pilots. We prefer to say so.
Do you need access to our prompts or our data?
No. The baseline reads cost and usage data (cost APIs, cloud billing, seat administration consoles), never the content of the conversations. When an API requires an admin key, extraction runs on your side and the key is not passed to us.
A baseline before scaling up
Thirty minutes to list your AI consumption channels and decide whether a pilot makes sense.