What is Cost per Task?
Cost per Task: The total cost of an AI system completing one unit of work, including model tokens, tools, infrastructure, and human review, compared to doing it manually.
Cost per task is the unit economics of an AI workflow. It adds up everything spent to complete one unit of work (one resolved ticket, one processed invoice, one answered call): model input and output tokens across every step, retries, embedding and search costs, third-party APIs, telephony, hosting, and the share of human review time. Divided by the number of completed tasks, it gives a number you can compare to the cost of doing the same task by hand.
Agents make this harder to predict because the number of steps varies per task. A simple request might take one model call; a messy one might take ten plus several tool calls. That is why you need per-run cost tracking, not just a monthly bill, and why outliers matter as much as the average.
Why it matters when you build or buy AI
If you do not know your cost per task, you do not know whether your automation is profitable. Get an estimate before the build and real numbers after launch, and watch them over time as volume, prompts, and models change. Prompt caching, model routing, and trimming context are the usual levers when the number is too high.
Example
An invoice workflow costs a few cents in model usage per document plus about one minute of review time on the 15 percent flagged for checking, compared to several minutes of manual entry per invoice before.
Related terms
More in Support & Ops
Book 20 minutes, or send a brief
A former CTO replies. You leave with a written plan and a fixed estimate.
Prefer LinkedIn? Connect with Harish.