The Metric
Total agent cost / number of successful resolutions = cost per resolution. For HubSpot outcome-priced agents, this is transparent. For others (token-billed), you compute it. It’s the honest measure of AI value.
Benchmark Ranges
Highly-automated internal workflows: $0.10–$0.50 per resolution. Customer-facing conversations: $0.50–$2.00. Complex multi-step workflows: $2–$10. Compare to human cost per equivalent resolution.
Optimizing the Metric
Improve resolution rate (better prompts, better grounding, better training data). Reduce cost (prompt caching, cheaper models where appropriate, reduced redundant calls). Each improvement moves the ratio.
Reporting Cadence
Monthly to ops leadership. Quarterly to CFO with trend analysis. When the metric degrades (cost up, resolution down), dig. When it improves, understand why — can you replicate?