Who Owns the AI Budget? A Practical Model for Team-Level Allocation
IT bought the licenses, engineering holds the API keys, finance pays the bill. An allocation model from one company budget down to every team.
Ask five people in a growing company who owns AI spend, and you will often get five answers. IT bought the Copilot licenses. Engineering holds the API keys. Product launched the customer-facing assistant. Finance pays the invoices. Everyone is involved, and nobody is accountable.
In FinOps, unowned spend is unmanaged spend. Here is a practical model for allocating AI budgets so that every dollar has an owner.
Start with one company budget
Before you can allocate, you need a single number. Set one company AI budget, monthly or yearly, that covers every channel: metered API usage, seat licenses, and fixed-price contracts with vendors that do not expose usage APIs.
That last category matters. If a contract sits outside the budget, teams will treat it as free, and your total will never reconcile with the general ledger.
Split by vendor, then allocate down the org tree
A workable allocation model has two dimensions:
- By vendor. How much of the company budget goes to each provider. This supports procurement, contract negotiations, and vendor concentration decisions.
- By organization. How the budget flows from the company to departments, from departments to teams, and from teams to individual users and agents.
The org dimension is where accountability lives. A department head owns a number. A team lead owns a slice of it. The structure should come from your identity directory, not from a spreadsheet someone maintains by hand.
Showback first, chargeback later
Most organizations should begin with showback: each department sees its own AI spend against its budget, with no internal billing. Showback builds the habit of looking at the number and creates the data history you need.
Chargeback, where costs are formally billed to cost centers, can follow once the allocation data is trusted. Moving to chargeback before teams trust the numbers usually creates disputes instead of accountability.
Rules that keep the model honest
An allocation model only holds if a few rules are enforced:
- No over-allocation. The sum of department budgets should never exceed the company budget. Block allocations that would break it.
- Default budgets for new users. New hires and new agents get a starting budget automatically, so nothing runs unallocated.
- Owners get the alerts. When a department reaches 80% of its budget, the department owner hears about it, not just FinOps.
- Usage, seats, and contracts in one basis. A team's spend includes its licenses as well as its tokens.
How KeepAir supports the model
KeepAir is built around this structure. You set one company budget, split it by vendor, and allocate it through your org tree down to department, team, and user, with live spend shown against every limit.
KeepAir imports people, departments, and teams from Microsoft Entra, or from your HR system through its Directory push API. Strict limits block over-allocation. Default budgets apply automatically to new users. Alert rules route every breach to the right owner in Teams, Slack, or email. For showback, every unit in the org tree shows spend by vendor, tokens, requests, seat cost, unused seats, savings potential, and budget.
The result
When every team can see its own AI spend against its own budget, the conversation changes. FinOps stops chasing explanations after the invoice arrives. Owners start managing their number during the month, the same way they manage headcount or cloud spend.
Give every AI dollar an owner.
Also published on LinkedIn. Follow KeepAir on LinkedIn