How to Calculate AI ROI: A Simple Framework for Your First Project

The most common mistake we see when companies start evaluating AI is the question they ask. They open the conversation with: "What is the ROI of AI?" That question has no answer. It is like asking what the ROI of software is. The question that has an answer is: "What is the ROI of automating this specific workflow for this specific team?"

When you narrow the question, the math becomes straightforward. You do not need a consultant or a spreadsheet model with thirty tabs. You need three inputs and twenty minutes.

The Three Inputs You Need

Before you build anything, before you talk to a vendor, before you write a business case, you need to answer three questions about the workflow you are targeting.

First: How many hours per week does this process take across your team? Not a rough estimate. A real number. Sit with the people doing the work and time it for a week. If four people each spend an average of five hours per week on a manual reporting process, that is twenty hours per week. Write it down.

Second: What is the fully-loaded hourly cost of the people doing it? Base salary is not the right number. You need to include benefits, payroll taxes, and overhead. A common rule of thumb is to multiply base salary by 1.3 to 1.5 to get the fully-loaded annual cost. Divide that by 2,080 working hours per year to get an hourly rate. For a team member earning $75,000 in base salary, the fully-loaded hourly rate lands somewhere around $48 to $54.

Third: What will the AI system cost to build and run annually? This includes the initial build cost, any third-party API costs, hosting, and ongoing maintenance. Get a real estimate from your implementation partner. Do not use the vendor's marketing page as your number.

The Math

Once you have those three inputs, the calculation is short.

Multiply hours saved per week by 52 to get annual hours saved. Multiply that by the fully-loaded hourly rate to get the annual labor value of the automation. That is your gross annual savings.

Subtract the annual operating cost of the AI system from that number to get your net annual savings.

To find the payback period, divide the one-time build cost by the net annual savings. The result tells you how many years it takes to recover the investment.

A Worked Example

Here is what this looks like with real numbers. Assume a team of four people each spend five hours per week on a manual reporting process. The fully-loaded hourly rate is $55. The annual labor cost of that process is 4 people times 5 hours times 52 weeks times $55, which equals $57,200 per year.

An AI system that automates this process costs $25,000 to build and $6,000 per year to maintain. In year one, the total cost is $31,000. Against $57,200 in labor savings, that is a net year-one gain of $26,200. The payback period on the build cost is under eight months.

In year two, the math gets better. The build cost is already recovered. The AI system costs $6,000 to run and saves $57,200 in labor. Net year-two savings: $51,200.

That is not a projection. That is arithmetic based on observable inputs.

What the Math Misses

The hours-saved calculation is almost always conservative. In our experience, the labor savings are the most visible part of the ROI, but they are not the whole story.

Three things typically go uncounted. First, error reduction. Manual processes produce errors. Errors produce downstream costs: rework, customer complaints, audit findings, and sometimes regulatory exposure. The cost of errors is rarely tracked, which means it rarely shows up in the ROI model. When you automate a process and eliminate the error rate, that value is real even if it is hard to quantify exactly.

Second, the work that was not getting done. When a team is spending twenty hours per week on a manual reporting process, there is other work they are not doing. That capacity was lost before the automation existed. After the automation, it comes back. Some of that recaptured time goes to higher-value work. Some of it means the team absorbs growth without adding headcount.

Third, consistency. An AI system does the same thing the same way every time. A team of four people doing the same process four different ways introduces variability that is costly to manage. Consistency has value, even when it is hard to put a number on it.

How to Pressure-Test the Number Before Presenting It

Before you take this model to your CFO, run it through three questions.

Is the hours estimate based on real observation or a guess? If you timed the process yourself, it is defensible. If you asked a manager what they thought their team spent, it is a guess. Guesses get picked apart in budget meetings. Real observations do not.

Have you accounted for implementation time and disruption cost? When a new system goes in, there is a period where productivity dips. People are learning something new while still managing their existing workload. That disruption period has a cost. If you have not accounted for it, your year-one savings are overstated.

What is the cost of NOT doing this? This question often does more work than the ROI model itself. If your team is spending 1,000 hours per year on a manual process and a competitor has automated that process, they are deploying those 1,000 hours elsewhere. The status quo has a cost. Make that cost visible.

What to Do Next

If you want a structured way to run this calculation for your specific workflows, our AI ROI calculator walks through the inputs and produces a shareable output you can bring to leadership. For a complete breakdown of what projects at each budget level include and exclude, see our AI transformation cost guide. If you want a broader view of where AI investments are likely to pay off across your operations, the AI competitive audit is the right starting point.