Insights
How do MSPs bill clients for AI usage?
MSPs bill clients for AI usage by metering consumption per client at the gateway or agent platform, pricing it on a published rate card that includes margin, adding a monthly "AI usage" line to each client's invoice with the detail attached, and booking the provider cost to cost of goods sold by client so gross margin is visible per account. It is the same model MSPs already use for cloud consumption and per-device monitoring, with one new problem: the provider bills you in aggregate, so attribution per client is your job.
The three billing models
| Model | How it works | Fits when |
|---|---|---|
| Pass-through plus margin | Metered tokens or requests at provider cost × markup (commonly 15 to 40%) | Client usage varies a lot; you want transparency |
| Bundled tiers | Fixed monthly fee per tier with an included usage allowance and overage rate | Predictable revenue; clients prefer a fixed number |
| Outcome or seat pricing | Per agent, per seat, or per completed task | You own the agent and the result, not just the plumbing |
Most MSPs start with pass-through plus margin because it is defensible and easy to explain, then move high-usage clients to tiers.
What you must have
- Per-client attribution. One API key or project per client per provider at minimum; a gateway that stamps every request with the client ID is better. Shared agents serving several clients need an allocation rule you can defend.
- A rate card the client has seen. Rates per million tokens by model, per request for tools, and the markup or margin stated plainly. Change it with notice, not mid-cycle.
- Usage detail with the invoice. A one-page statement per client: agents, tokens, requests, rate, amount, prior month. Disputes drop when the detail is attached.
- Cost booked by client. Provider invoices coded to COGS with a client segment, so each account's margin on AI is a report, not a guess.
- Terms for agent-initiated spend. If a client's agent buys a tool or data on its own through your platform, the agreement has to say whose cost it is and at what markup.
Worked example
Provider bill to the MSP for September: $9,600. Gateway attribution by client:
| Client | Provider cost | Markup | Billed |
|---|---|---|---|
| Harbor Dental Group | $3,200 | 30% | $4,160 |
| Westfield Logistics | $4,800 | 25% | $6,000 |
| Oak Street Law | $1,600 | 30% | $2,080 |
| Total | $9,600 | $12,240 |
Client invoice line, Harbor Dental Group
| Item | Detail | Amount |
|---|---|---|
| AI usage, September | Intake agent 2.1B input / 410M output tokens; see attached statement | $4,160.00 |
MSP's books, September
| Account | Debit | Credit | Client |
|---|---|---|---|
| Accounts receivable | 12,240 | (by client) | |
| AI usage revenue | 12,240 | (by client) | |
| COGS: AI provider cost | 9,600 | (by client) | |
| Accrued AI provider cost | 9,600 |
Gross margin on AI for the month: $2,640, visible per client from the segments.
Mistakes that cost margin
- Billing on the provider invoice. It arrives after your client invoices go out and lumps periods. Bill metered usage; reconcile provider cost separately.
- No cap or alert per client. A runaway agent loop can burn a month's margin in a weekend. Set a budget per client with alerts to you and to them.
- Hiding the markup. Clients find out. A stated margin survives procurement; a hidden one does not.
- One shared key for all clients. You cannot attribute, so you cannot bill accurately or defend a dispute.
Related
How do I do chargeback for AI usage? · Is there an open standard for AI agent spend data?
Worked examples are illustrative. Markup levels vary by market; client names are fictional.
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.