How to estimate the ROI of an AI agent

Short answer: estimate the return on an AI agent in hours, not in promises. Multiply the people doing the work by their weekly hours on it, by the share an agent can realistically take, by 48 working weeks. Then price those hours at your loaded cost per hour. Treat the result as capacity returned to the team, not headcount removed, and subtract what the agent costs to run.

How do you estimate the ROI of an AI agent?

Start with one workflow, not the whole department. ROI estimates fail when they cover everything and commit to nothing. Pick a task that is repetitive, rule-based and done by several people, then use four inputs:

People who do the work today.
Hours per person each week spent on it.
Cost per hour, fully loaded with salary, benefits and overhead.
Share the agent can take on, as a percentage of those hours.

Capacity returned: the hours an agent frees up for a team to spend on review, exceptions and higher-value work. It is a different thing from a saving, which only exists if you spend less.

The formula is: people × hours per week × 48 weeks × share = hours returned per year. Multiply by cost per hour for the value of that capacity. Divide by 1,800 hours to express it as full-time equivalents.

A worked example: invoice processing

Fictional team and numbers, chosen to show the method. Six people each spend 15 hours a week processing invoices, at a loaded cost of $65 an hour. An agent can take on 60% of that work.

Step	Calculation	Result
Hours on the work each week	6 people × 15 hours	90 hours
Hours the agent handles each week	90 × 60%	54 hours
Hours returned per year	54 × 48 weeks	2,592 hours
Value of that capacity	2,592 × $65	$168,480
In full-time equivalents	2,592 ÷ 1,800	1.4 FTE

Notice what the team keeps: 36 hours a week for approvals, exceptions and the invoices that need judgement. The number is real, but it is capacity. Whether it becomes a saving depends on what the team does with it: absorbing growth without hiring, shortening the close, or taking on work that was waiting.

Return on an agent is hours given back, and the business decides what they are worth.

What should you subtract from the return?

The cost of running the agent, and the cost of keeping people in the loop. A fair estimate nets out three things:

Run cost. Model usage is metered, so it scales with volume. Our guide to AI agent costs shows how to forecast it with four numbers, and the AI Budget Planner does the arithmetic.
Build and change effort. Someone has to set the agent up, test it and maintain it as your process changes.
Review time. The hours people spend checking the agent’s work. This is why the share should not be 100%.

Because you want a person on every key decision. We recommend capping automation at around 80% for most finance and operations tasks, and often lower to start. An agent that handles 60% of invoice work, with people approving exceptions and anything above a threshold, is both safer and easier to trust than one handling nearly everything.

A lower share also makes the estimate more honest. If the case only works at 95%, it is not a strong case.

What makes an ROI estimate credible?

Use your own hours. Ask the people doing the work, or sample a week, rather than relying on benchmarks.
Test the share. Show the result at 40%, 60% and 80%. If the project only pays at the top, say so.
Name where the capacity goes. A returned hour with no plan behind it tends to disappear.
Revisit after launch. Compare the estimate with measured hours at 30, 60 and 90 days.

Key takeaways

Estimate one workflow in hours: people × weekly hours × 48 × share.
Call the result capacity returned, and decide what the team will do with it.
Subtract run cost, build effort and review time.
Keep the share below about 80% and keep a person on key decisions.
Show a range, and check the estimate against measured hours after launch.

Try your own numbers

The Agent ROI Calculator takes the four inputs, shows hours returned, capacity value and full-time equivalents, and splits a working week between the agent and your people. It is free, runs in your browser, and starts from presets for common workflows.

Frequently asked questions

How do you calculate the ROI of an AI agent?

Multiply people by weekly hours on the task, by 48 working weeks, by the share an agent can take on. Price those hours at your loaded hourly cost, then subtract run cost, build effort and review time.

Is AI agent ROI a saving or capacity?

Mostly capacity. Hours returned only become a cost saving if you spend less, for example by not hiring. Otherwise they become faster cycles or room for new work, which is still valuable but should be named.

What share of a task can an AI agent take on?

It depends on the task. As a planning rule, we suggest staying at or below roughly 80% and keeping a person on approvals and judgement calls. Start lower, and raise the share as measured quality earns trust.

What should be included in the cost of an AI agent?

Metered model usage, the effort to build and maintain the agent, and the time people spend reviewing its work.