# Agent cost allocation model

Use this model to attribute AI costs to the workflows, departments, customers, and outcomes that create them. Without allocation, AI spend becomes a shared infrastructure line item that is hard to govern and harder to optimize.

## Cost dimensions

- Runtime: model tokens, embeddings, reranking, tool calls, orchestration steps, and retries.
- Operations: monitoring, evaluation runs, human review, support escalations, and incident response.
- Delivery: integration build, prompt changes, data preparation, red-team testing, and rollout enablement.
- Business unit: department, workflow, user segment, customer tier, and initiative owner.
- Outcome: resolved task, avoided manual work, reduced cycle time, improved compliance, or quality gain.

## Review cadence

Review cost per successful outcome monthly. Investigate spend growth that is not matched by adoption, quality, risk reduction, or revenue impact.