Cost Per Action (CPA)
The total cost for an AI agent to complete a single meaningful action—such as resolving a support ticket, qualifying a lead, processing an invoice, or generating a report. CPA includes LLM inference costs, tool/API call costs, and infrastructure overhead. CPA is the primary ROI metric for comparing AI agents against human labor or alternative automation: if a human resolves a support ticket for $7 and an AI agent does it for $0.35, the 20x cost reduction justifies the investment.
Written by Max Zeshut
Founder at Agentmelt · Last updated Sep 9, 2026
Example
A support team calculates their AI agent's CPA: average ticket uses 3,000 input tokens ($0.009) + 800 output tokens ($0.012) + 2 knowledge base queries ($0.002) + CRM API call ($0.001) = $0.024 per ticket. Their human agents cost $8.50 per ticket including salary, tools, and overhead. The AI agent handles 60% of tickets autonomously at $0.024 each.
Frequently asked questions
- How do I calculate CPA for my AI agent?
- Sum all costs per completed action: (1) LLM API costs (input + output tokens × price per token), (2) embedding/retrieval costs (vector search queries), (3) external API costs (CRM writes, email sends), (4) infrastructure costs (hosting, compute). Divide by the number of successfully completed actions—don't count escalated or failed attempts. Compare against your current cost per action to calculate ROI.