A better way to measure AI costs starts with the work, not the bill

AI bills show what an agent consumed, but not whether the work was worth the cost. Hanah-Marie Darley, co-founder and chief AI officer of Geordie, argues that businesses should trace spending to the task, the agent’s behaviour and the outcome.

Watch Desk analysis

What happened

In a Forbes Technology Council essay, Darley sets out a way to assess the economics of AI agents by following a piece of work from its cost to its result. That means looking beyond model invoices and token counts to the agent’s task, the models and tools it used, retries, data access and any human review.

Her suggested sequence is spend, agent, work, behaviour, resources and outcome. She argues that value should be judged in terms relevant to each task, whether that is time saved, accuracy, throughput or work a team could not otherwise take on.

Why it matters

The same bill can describe very different outcomes. A premium model may be a sensible choice for a valuable, time-sensitive decision, but harder to justify for routine classification. Without context about what an agent was asked to do and what happened, a higher cost is just a number with expensive-looking shoes.

Darley recommends starting with a small number of important workflows, defining the intended outcome and naming who is responsible for deciding whether the resources were justified. Teams can then make the work traceable and use the evidence to expand valuable workflows or improve ones where costs outweigh results.

Our read

This is a useful management framework, not a universal formula for return on investment. Its strength is making AI spending answerable to the job it was meant to do, rather than asking a token tally to moonlight as a business case.

For teams already using agents, choose one consequential workflow and record its intended result, costs, tools, retries and human input. That gives the next budget conversation something better than a hunch and a colourful dashboard.

What to watch

• Whether teams can link agent costs to specific tasks and outcomes.
• Which measures different teams use to judge whether a workflow is worthwhile.
• Whether traceability leads to changes in model choice, tool use or task design.

Discussion spark: Should businesses judge AI agents by a common return-on-investment measure, or should each team define value for its own work?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.