Tool
What does your AI agent cost per resolved ticket?
Token counts are a vanity metric. This works out the number finance understands, then shows what caching, resolution rate and model routing do to it. Prices are editable so it stays useful when they change.
How it's calculated
Input cost per conversation is prompt tokens per turn × turns × a blended input price, where the blended price is the normal input price for the uncached share and the cached price for the share of the prompt that is both a stable prefix and actually served from cache. Output cost is output tokens × turns × output price. Cost per resolved ticket divides monthly spend by conversations × resolution rate.
The model comes from running the support agent on REConnect, where fixing three cache invalidators cut input cost by more than half. The reasoning is in Measure AI features by cost per resolved task and Prompt caching in production.
Want this done for your system, with real usage logs and an eval set? Get in touch. Built by Zain Akram.