AI Glossary · Letter C

Cost Per Task.

The full cost of getting one finished piece of work out of an AI system, including retries, reasoning, and tool calls, which is the number that actually matters for pricing and budgeting AI work.

Also known as Cost per outcome, unit cost of AI work

What it is

A working definition of Cost Per Task.

AI vendors publish prices per million tokens, which makes models look easy to compare. But a task rarely costs one prompt and one answer. Reasoning models spend extra tokens thinking. Agents call tools, read files, retry when a step fails, and carry long context forward. Some vendors charge more once a prompt passes a size threshold.

Cost per task adds all of that up for one unit of finished work: one approved banner set, one weekly report, one competitor scan. Two models with similar token prices can differ several times over on cost per task, and a more expensive model sometimes wins because it gets the job right on the first try.

Why ad agencies care

Why Cost Per Task matters in agency work.

It is the bridge between an AI invoice and a client scope.

Pricing needs a unit. Agencies pricing AI-assisted deliverables cannot quote in tokens. Knowing that a localized ad set costs, say, a few dollars in compute versus a few cents tells the agency where AI changes the margin and where it barely registers.

It exposes hidden waste. Long agent runs that loop, retry, or reread the same documents can cost far more than anyone expects. Measuring per task surfaces the workflows worth fixing.

It makes model choice concrete. Instead of arguing about which model is better, teams can run the same task on two models and compare cost per approved output.

In practice

What cost per task looks like inside a working ad agency.

An agency’s automation lead builds an agent that turns a master video script into twelve market versions. The token price looks trivial, but when she logs a week of runs she finds the average job costs far more than expected because the agent rereads the full brand book on every step and retries translations that fail a length check. She caches the brand book, tightens the length instruction, and routes the first draft to a cheaper model. Cost per localized script falls by more than half, and the agency can now quote localization as a fixed line item.

Build AI workflows that actually run through The Creative Cadence Workshop.

The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.