Blog

How to measure AI agent profitability

AI agent profitability is whether the revenue attached to agent work covers the variable delivery cost of producing successful outcomes. Measuring it requires both commercial and delivery data.

· Arbiter Margin

Define the unit you sell

Agents may be sold as seats, usage, or outcomes. Profitability measurement should use the same unit your contract uses, then map delivery cost onto that unit.

Measure cost per task and per successful outcome

Cost per AI agent task includes failed and retried work. Cost per successful outcome divides delivery cost by completed results. Outcome pricing lives or dies on the second metric. A labeled walkthrough is in how to calculate AI cost per completed task.

Example (synthetic)

Example only. An agent workflow completes 200 successful outcomes in a week. Delivery cost attributed to that workflow is $1,000. Cost per successful outcome = $5.

If the customer pays a usage package that effectively yields $4 per outcome for this mix, the workflow is unprofitable even if the account’s seat fee looks healthy in isolation.

Roll up to customer and portfolio

Agent workflows should roll into customer gross margin and then portfolio views. That is how AI unit economics informs packaging and AI agent pricing strategy.

FAQ

Is token spend enough to measure agent profitability?
Usually not. Tools, retries, and multi-step planning can dominate. Attribute the full delivery path when possible.
How often should we review agent margins?
Often enough to catch mix shifts — many teams review weekly for new agents and monthly for stable ones.