# How do you measure the ROI of an AI agent?

Count 4 numbers. The hours the task takes today, priced at what those hours cost. The share of that work the agent takes over in weekly use. What it costs to run each month. What it cost to build. The yearly value, minus the yearly run cost, divided by the build cost, gives the ROI.

By Surinder Thakur, AI Product Designer and Builder. Published 17 Nov 2026. Topic: [Building an AI tool](https://www.surinder.design/work?type=building-an-ai-tool).

## Why start with today's cost?

Because the agent can only save what the work costs now. List who does the task, how often, and how long it takes.

Price those hours at what the company pays for them. That is the ceiling on what the agent can return.

## Why does weekly use matter more than accuracy?

Because a tool nobody opens saves nothing. An agent can be right almost every time and still sit unused beside the old spreadsheet.

After launch, track 1 number: the share of the task the agent handles each week. Savings follow use, never the launch date.

## What goes into the cost side?

2 costs, kept apart. The build is paid once. The run cost comes every month. It covers model usage, hosting and upkeep.

Leave the run cost out and the ROI looks better on paper than in the bank.

## When should you measure it?

Before the build as an estimate, then every week after launch as a fact. The estimate decides whether to build at all.

In the 2 week Blueprint, the estimate becomes a business case in numbers. If it says no, you stop before the build spend.

## In 1 picture

4 numbers make the ROI of an AI agent, measured before and after the build.

1. Today's cost: Hours on the task, priced
2. Weekly use: Share the agent takes over
3. Minus run cost: Model, hosting, upkeep
4. Over build cost: Paid once
5. ROI: Measured every week

Savings follow use, never the launch date.

## Next questions

- [What does an AI agent cost to run each month?](https://www.surinder.design/work/what-does-an-ai-agent-cost-to-run): 3 parts, paid every month. The model bill, which grows with every task the agent handles.
- [How do you know if an AI agent is actually working?](https://www.surinder.design/work/how-to-know-if-an-ai-agent-is-working): Read 4 numbers every week. Tasks it finished without a person.
- [What if the build does not pay off?](https://www.surinder.design/work/what-if-the-build-does-not-pay-off): Then you find out in the Blueprint, before the build spend.
- [Which task in my business should get an AI agent first?](https://www.surinder.design/work/which-task-should-get-an-ai-agent-first): Pick a task that repeats every week, is easy to check, and has 1 owner.

## Questions that lead here

- [How do you pick an AI agent project that will not get cancelled?](https://www.surinder.design/work/how-to-pick-an-ai-agent-project): Pick a job with a known cost, clear rules and 1 owner.

Page: https://www.surinder.design/work/how-to-measure-the-roi-of-an-ai-agent
