Insights
How do I do chargeback for AI usage?
Chargeback for AI usage means metering each agent's consumption, attributing it to the department or customer the agent worked for, pricing it at an agreed internal rate, issuing a monthly statement to each owner, and posting the entries that move the cost from a central holding account to the consuming cost center. Showback is the same process without the entries; chargeback puts the cost on the owner's P&L.
Showback first, then chargeback
Start with two or three months of showback. Owners see their statement, dispute attribution errors, and the rate card gets fixed before money moves. Switching to chargeback after owners trust the numbers avoids the fight over the first invoice.
What you need in place
- Attribution at runtime. Every usage record carries agent identity and a cost object. Shared agents (an internal assistant used by many teams) need an allocation rule: by request count, by tokens, or by seats.
- A rate card. Either pass-through at provider cost or a blended internal rate that includes platform overhead. Publish it. Change it on a schedule, not mid-month.
- A statement format. One page per owner: usage by agent, rate applied, amount, prior-month comparison, and a link to the detail. Finance recognizes a statement; nobody reads a dashboard at close.
- A dispute window. Five business days after the statement before entries post.
- The entries. Central cost in, departmental cost out.
Worked example
Central AI platform account incurs $12,000 in September across a shared assistant and two dedicated agents.
Attribution
| Cost object | Agent | Basis | Amount |
|---|---|---|---|
| Sales | Lead-research agent (dedicated) | Direct | $5,100 |
| Customer Success | Support agent (dedicated) | Direct | $4,300 |
| Finance | Shared assistant | 40% of requests | $1,040 |
| HR | Shared assistant | 35% of requests | $910 |
| Legal | Shared assistant | 25% of requests | $650 |
| Total | $12,000 |
Statement line for Sales
| Agent | Input tokens | Output tokens | Rate basis | Amount | vs. Aug |
|---|---|---|---|---|---|
| Lead-research agent | 610M | 95M | Pass-through | $5,100 | +12% |
Chargeback entry, October 7 (after dispute window)
| Account | Debit | Credit | Cost center |
|---|---|---|---|
| AI services expense | 5,100 | Sales | |
| AI services expense | 4,300 | Customer Success | |
| AI services expense | 1,040 | Finance | |
| AI services expense | 910 | HR | |
| AI services expense | 650 | Legal | |
| AI services expense (central clearing) | 12,000 | IT Platform |
The central account nets to zero each month. If it does not, the attribution is incomplete and the residual stays visible rather than being smeared.
Common failure modes
- Allocating by headcount. It is easy and wrong; a ten-person team can run an agent that costs more than a hundred-person team's usage.
- Charging back the invoice instead of the usage. Invoices arrive late and lump periods together. Charge back metered usage, true up on invoice.
- No owner for agent-initiated purchases. If an agent buys a tool or data on its own, the purchase inherits the agent's cost object. Decide that rule before the first one happens.
Related
How do I account for AI agent spend in the general ledger? · Who owns AI spend, finance or IT? · How do MSPs bill clients for AI usage?
Worked examples are illustrative. Allocation policy should be agreed with department owners and your controller.
Not sure which of your AI costs are being booked? Run the free Agent Spend Assessment.
Onaro Meridian is FinOps for agentic AI: the system of record that attributes, controls and books what AI agents spend.
Brian Diamond
Brian Diamond is a fractional Chief AI Officer and founder of Onaro. He has spent 30 years running infrastructure operations and founded LANStatus, a Connecticut managed services provider and Microsoft partner, in 2001. He holds a Chief AI Officer certification and writes the CAIO Brief on AI leadership for finance and operations.