How to Calculate AI ROI: The Simple Formula Every Business Owner Needs

It matters, because the evidence cuts both ways. MIT NANDA's 2025 study found 95% of organizations reporting no measurable return from generative AI so far. Google Cloud's 2025 survey of 3,466 senior leaders found 74% saying they'd seen a return within the first year. Both can be true. The difference is usually whether someone did the math before buying.

The formula is the easy bit. The hard bit is being honest about what goes into each side.

A complete cost picture has four lines. The first is on the invoice. The other three are your team's time, and they're the ones people leave out.

• The tool or service itself. Subscription, setup fees, any integration work.
• Training. A few hours per person who'll use it, once. Multiply by their hourly cost.
• Data cleanup. If your client list has three versions of every company, someone has to fix that first.
• Oversight. Someone checks a sample of the AI's work, every week at first, then monthly. That's real time.

AI can pay off in four ways. Some are easy to count, some aren't, and one is easy to fool yourself with.

• Time saved. Hours per week × loaded hourly cost × working weeks. For example, 15 hours × $45 × 48 weeks = $32,400 a year.
• Fewer errors. Invoices sent twice, a missed renewal, data typed into the wrong field. Count the ones you can actually name from last year.
• More throughput. Serving more clients with the same team. Count it only if you have the demand.
• Retention. People who don't spend their week on copy-paste are less likely to quit. Real, but hard to put a number on, so leave it out of the math and treat it as a bonus.

This is an illustration with round, conservative numbers, not a client result. Swap in your own.

Say a 12-person marketing agency hands its repetitive admin (inbox triage, invoice entry, meeting scheduling and status reminders) to one AI employee-equivalent. Four staff currently share that work.

• AI employee: $21,500 a year (QLIA's price for an AI employee-equivalent)
• Training: 4 people × 6 hours × $45 an hour = $1,080
• Data cleanup and setup: one week of a staff member's time, 40 hours × $45 = $1,800
• Oversight: 3 hours a month of the owner's time × 12 months × $75 = $2,700
• Total cost: $27,080
• Time saved: 20 hours a week across the four staff × $45 × 48 weeks = $43,200
• Errors avoided: $3,000 (a handful of double-billed or missed invoices a year)
• Total gain: $46,200

Run it through the formula. ROI = ($46,200 − $27,080) ÷ $27,080 × 100 = about 71%. Monthly gain is $46,200 ÷ 12 = $3,850, so payback = $27,080 ÷ $3,850 = about seven months.

Not a moonshot. That's the point. A believable 71% you can defend to your accountant beats a 900% projection nobody trusts. And year two usually looks better, because the setup and training costs don't repeat.

Now the trap from above. If those 20 hours a week don't go anywhere useful, the $43,200 isn't real. Before you sign, decide where the hours go: client work, a hire you won't need, or the overtime you'll stop paying.

If six months in you can't point to a measurable gain, don't panic and don't double down. Check four things:

• Wrong job. You automated something that happens twice a month. Move it to something that happens every day.
• Bad data. The AI is working from messy records. Clean the source, not the output.
• Nobody's using it. Staff went back to the old way. Find out why; usually it's one annoying step that can be fixed.
• Wrong measure. You're counting logins instead of hours saved or invoices processed.

Take the four cost lines and the gains you can actually name, and put them through the formula. If the answer is thin, that's useful too. It means you picked the wrong first job, and it cost you an afternoon instead of a year.

Or start with our free AI ROI calculator. Answer a few questions about your team and your repetitive work, and it gives you an estimate you can check against this formula.