Know who is spending what.
Attribution that holds up in a budget review. Cost by team, project, workflow, product, department and customer, from the same source of truth.
How is AI spend allocated to teams and customers?
Camaze maps every cost source once, using the identifiers each provider already exposes: API keys, projects, workspaces, cloud tags, seat assignments and request metadata. Those map to your own org structure, so every subsequent charge is attributed automatically. Spend that cannot be attributed is reported as unattributed rather than distributed silently, and the result supports showback, chargeback, and a per-customer cost to serve.
Everyone agrees AI is expensive. Nobody owns the number.
An unallocated cost has no owner, and a cost with no owner does not get managed. It gets discussed.
More detail
The engineering team assumes finance is watching it. Finance assumes the team that set it up is watching it. The line grows, and the first person to genuinely look at it is whoever is preparing the board pack.
Meanwhile the questions that matter cannot be answered at all. Which product line does this belong to. Is the AI cost of serving an enterprise account eating the margin on it. Should this be capitalized, and against which project.
- Provider bills organized by API key, which nobody can map back to a team
- One shared key across four teams, so the split is a guess
- AI seats assigned centrally with no cost back to the department using them
- Self-hosted GPU spend sitting in a shared infrastructure account
Illustrative product view. Figures are examples.
Map it once. Everything after that is automatic.
Allocation fails when it depends on people remembering to tag things. Camaze derives attribution from identifiers that already exist and already have to be correct for the system to run.
- Keys, projects, workspaces and cloud tags mapped to your org structure
- Seat assignments mapped to departments, with actual usage attached
- Request metadata used for workflow, feature and customer attribution where available
- Shared and platform costs split by a rule you choose, applied consistently
- Unattributed spend reported openly, with the reason it could not be placed
Give each team its own number.
Showback changes behavior before any money moves.
More detail
When a team can see its own AI spend, its trend, and how it compares to the budget it agreed to, the conversation stops being about AI in general and starts being about a specific workflow.
Each owner gets their own view and a weekly digest. Finance gets the consolidated position. Both are reading the same figures.
- A per-team view with budget, actual, forecast and variance
- Weekly digest to the owner, in Slack, Teams or email
- Drill from a team total to the workflow and model driving it
- Month-on-month change, with the largest movements listed first
| Project | Model | Share | Cost | Change |
|---|---|---|---|---|
| Support automation | Claude Sonnet | $34,800 | +22% | |
| Search and ranking | GPT-4o mini | $21,400 | -6% | |
| Sales copilot | GPT-4o | $18,900 | +31% | |
| Docs assistant | Llama 3.1 70B, self-hosted | $15,200 | +9% | |
| Data enrichment | Gemini 1.5 Flash | $11,600 | -14% | |
| Internal agents | Mixed | $9,800 | +64% | |
| Engineering seats | Cursor, Copilot | $8,300 | 0% |
Illustrative product view. Figures are examples.
Move the cost to the department that caused it.
When AI spend is charged back, it enters the same discipline as every other cost.
More detail
It appears in a departmental P&L, it is defended in a planning cycle, and it is owned by someone with a budget.
Camaze produces the allocation, the supporting detail, and a journal-ready export that matches your chart of accounts. Shared costs are split by whichever rule you select, and the rule is applied identically every period so the figures reconcile.
- Journal-ready exports mapped to your chart of accounts
- Configurable split rules for shared platform and self-hosted costs
- A full audit trail from a journal line back to the underlying usage
- Restatement handled cleanly when a mapping changes mid-period
| Report | Runs | Lands in |
|---|---|---|
| Monthly close pack | 1st of the month | Excel, emailed to finance |
| Chargeback journal | 1st of the month | ERP export, journal-ready |
| Board AI summary | Quarterly | PDF and slides |
| Cost by team | Every Monday | Google Sheets, live |
| Raw cost and usage | Nightly | Snowflake |
| Budget variance digest | Every Friday | Slack, #finance |
Illustrative product view. Figures are examples.
What AI costs to serve one customer.
For any company selling an AI-backed product, this is the number that decides whether the business works.
More detail
AI is a variable cost that scales with usage, which means it behaves like cost of goods sold and belongs in gross margin, not in a general technology line.
Camaze reports AI cost per account, per segment and per plan, next to the revenue from that account. Heavy users become visible before they become a margin problem, and pricing decisions get made on evidence.
- AI cost per account, per segment and per pricing plan
- AI as a share of revenue, and as a component of gross margin
- The accounts whose AI cost is growing faster than their contract value
- Cost per feature, for deciding what is included and what is metered
| Segment | Revenue per account | AI cost per account | AI as share of revenue |
|---|---|---|---|
| Enterprise, tier 1 | $18,400 | $2,180 | 11.8% |
| Enterprise, tier 2 | $9,200 | $1,640 | 17.8% |
| Mid-market | $3,100 | $291 | 9.4% |
| Growth | $980 | $142 | 14.5% |
| Self-serve | $79 | $31 | 39.2% |
Illustrative product view. Figures are examples.
See your spend attributed.
Bring your team structure to a 30 minute session and we will show you what allocated AI spend looks like against it.
Getting to allocated spend
Most companies reach reliable attribution in the first two weeks.
Connect and observe
Sources connect read-only and history backfills. At this point spend is visible by provider and by key, and largely unattributed. That is the honest starting position.
Map to your structure
Keys, projects, workspaces, tags and seats are mapped to your teams and cost centers. This is a working session, usually an hour or two, and it is the only manual step.
Publish and charge back
Owners receive their own view and digest. Finance runs showback for a period or two to settle disputes, then switches on chargeback and journal export.
Keep reading
Spend visibility
The collection layer underneath allocation: every provider, model and self-hosted workload on one cost basis.
Read moreCost per customer
What AI adds to the cost of serving one account, and which accounts are eating their margin.
Read moreAI budget ownership
Giving the AI line an owner, a budget and a review cycle instead of a shared worry.
Read moreQuestions people ask
What if several teams share one API key?
That is the normal starting state. Where request metadata is available, spend can be attributed below the key. Where it is not, Camaze applies a split rule you define and reports the number as apportioned rather than measured, so nobody mistakes an estimate for a fact. Most companies then split the key, because the estimate is uncomfortable enough to prompt it.
Can we allocate self-hosted GPU spend?
Yes. GPU compute is attributed to the endpoints and models it serves, and those inherit the workload and team mapping. Idle capacity is held separately as a platform cost rather than charged to whichever team happened to use the endpoint last.
Does this support chargeback into our ERP?
Yes. Allocation output is available as a journal-ready export mapped to your chart of accounts, on your close schedule. See reporting and exports.
What happens to costs nobody owns?
They are reported as unattributed, with the reason. Unattributed spend is a signal, not an error. It usually means a new key, a new workflow or a tool that has not been mapped yet, and it is often where the first savings finding comes from.
Can we allocate to customers if we do not tag requests?
Partially. Without per-request customer identifiers, Camaze can allocate down to workflow and feature, then apportion to customers using a usage driver you already have, such as seats or API calls. Adding a customer identifier to requests upgrades this from apportioned to measured, and it is usually a small engineering change.