How to calculate the ROI of an AI system before building it

"How much will it save me?" is the right question to ask before investing in artificial intelligence. And while many AI projects start out of enthusiasm, the ones that survive the next budget review are the ones that can answer it.

The good news is that the formula for return on investment doesn't change just because it's AI. What changes is what counts as a benefit and what counts as a cost.

The formula

ROI compares what you gain with what you put in:

ROI = (total benefit − total cost) ÷ total cost

If the result is positive, the project returns more than it cost. Along with ROI, it is useful to calculate the payback period: how many months it takes until the cumulative benefits equal the investment. A project with good ROI that takes three years to pay for itself may not be a priority compared to a more modest one that pays for itself in months.

What counts as a benefit

The benefits of an AI system usually fall into four groups:

Benefit type	How it is calculated	Example
Time saved	Hours freed up × hourly cost of the person doing them	Document review, reconciliations, reports
Errors avoided	Errors that stop occurring × cost of each error	Fines, rework, returns
Additional revenue	Sales that previously didn't happen	Faster response to quotes, better commercial follow-up
Capacity	Additional volume you can handle without hiring	More requests processed with the same team

Two precautions when calculating time saved:

1. Freed-up hours aren't money until they are spent on something else. If the team saves hours but doesn't dedicate them to something of higher value, the benefit exists on paper and not on the income statement. Define from the start what people will do with that time.
2. Adoption is not one hundred percent. Multiply the benefit by the realistic percentage of cases where the system will actually be used. An excellent system that half the team ignores delivers half the return.

What counts as a cost

This is where calculations often fall short. The cost of an AI system isn't just building it:

• Implementation: design, development, integration with your systems, and testing.
• Data preparation: cleaning, defining, and connecting the information. This is often the longest part.
• Model usage: models are billed by consumption (by volume of processed text) or require their own infrastructure if they run on your servers.
• Infrastructure: cloud or servers, storage, and security.
• Operation and maintenance: monitoring, adjustments when processes change, model updates.
• Training and change management: the time your team spends learning to work with the system.
• Internal time: your experts' hours during the project, explaining processes and validating results.

Separate one-time costs (implementation, data preparation) from recurring ones (model usage, infrastructure, maintenance). The former define the investment; the latter determine whether the monthly benefit is enough to sustain the system.

A five-step method

1. Choose a specific process, not "using AI in the company." ROI is calculated per use case.
2. Measure the baseline: how much time it takes today, how many people participate, how many errors occur. Without a baseline, no comparison is possible.
3. Estimate the benefit with explicit assumptions: what percentage of the work the system takes on and what adoption you expect. Write them down so you can review them later.
4. Add up all costs, one-time and recurring, over a one- to three-year horizon.
5. Calculate ROI and payback period, and repeat with a pessimistic scenario. If the project only works in the optimistic scenario, it's not ready.

The most common mistake: calculating the ROI of the wrong idea

Most companies have more AI ideas than they can execute, and they often pick the flashiest one instead of the one with the highest return. Calculating the ROI of a project properly is of little use if it was the wrong project.

That's why, before calculating, it's best to compare. List the ideas from the different areas, estimate the impact and difficulty for each, and calculate the detailed ROI only for the two or three most promising ones.

Summary

AI ROI is calculated just like any investment, with two caveats: count all costs, not just development, and adjust benefits for actual adoption. Measure the baseline before starting, write down your assumptions, and compare several ideas before investing in one.

Start by knowing which idea is worth it

Comparing ideas across the entire company is precisely what IRIS, our diagnostic tool, does. An AI agent interviews your entire team and, out of all their ideas, shows you which one will save the most time and have the greatest impact. That way, you calculate the ROI of the right project from the start.

If you'd like to learn more, schedule a call with our team.