Resources · Getting started
AI ROI calculator: the formula, the inputs, and what the number leaves out
Updated 2026-11-03
The formula, in one line
Every AI ROI calculator worth using runs the same arithmetic, whether it shows you the formula or hides it behind a slider:
Yearly value = hours per week × 52 × loaded cost per hour × share automated
Then the part most calculators skip:
Return = yearly value − one-time build cost − yearly running cost
That is the whole thing. A task that takes four hours a week at a loaded cost of 40 dollars an hour is 8,320 dollars a year of time. If software can take 60 percent of it, the value is just under 5,000 dollars a year. If the automation costs 3,000 dollars to build and a few hundred a year to run, it pays for itself inside the first year and keeps paying after. If it costs 12,000 dollars to build, it does not, and you should not do it.
We built a free calculator that runs the first line for you, in your browser, with no email required. This page is the part a calculator cannot do: picking honest inputs and reading the result.
The three inputs that decide everything
1. Hours per week
Owners guess this low, almost every time. The task feels like ten minutes because each instance is ten minutes. Count the instances. Chasing one quote is a few minutes. Chasing twenty quotes a week, with the follow-up and the follow-up to the follow-up, is an afternoon.
The honest way to get the number is to ask the person who does the work, then watch for a week. Write down each time the task comes up and how long it took. The total is usually larger than the guess, sometimes by half, and that gap is the difference between a project that looks marginal and one that obviously pays.
Only count the repetitive slice. The thirty minutes of judgment at the end of a task is not automatable, so leave it out of the hours or it will inflate the result.
2. Loaded cost per hour
Use what the hour actually costs the business, not the hourly wage. Loaded cost is wage plus payroll taxes, benefits, paid time off, and the share of rent and software the seat carries. A common rule of thumb is 1.2 to 1.5 times the wage, though your bookkeeper can give you the real multiplier from your own payroll.
For an owner, the wage is the wrong number entirely. Use what an hour of your attention is worth to the business. If you spend six hours a week on invoicing that could have gone to sales calls or estimates, the cost of those six hours is the sales you did not make, and that is usually a much bigger number than anything on a pay stub.
3. Share automated
This is the slider everyone pushes to the right. Keep it honest. Repetitive office work with clear rules, like quote follow-up, invoicing and reminders, review responses, report assembly, or moving data between two systems, usually gives back 30 to 80 percent of the time. The rest is review, exceptions, and the cases where someone needs to pick up the phone.
Our own calculator caps the slider at 80 percent on purpose. Some of every job stays human. A calculator that lets you type 100 percent is helping you fool yourself, and the number you get will not survive contact with the first month of real use.
If you are not sure which tasks even qualify, the AI agents guide draws the line between work that is rules plus volume and work that is judgment plus one-offs. Only the first kind belongs in the calculator.
The cost side, which most calculators leave blank
A yearly value with no cost next to it is not an ROI, it is a wish. Put real numbers against it.
One-time build. For a focused automation, think low thousands, scoped from a written assessment. We lay out the pieces in what AI costs a small business. If you are doing it yourself with an off-the-shelf tool, the build cost is your own hours, which you should price at the same loaded rate you used above.
Yearly running cost. Model usage for a normal small business volume is often under a hundred dollars a month. Add any subscription the automation depends on, and add a line for upkeep: someone has to check it, fix it when a vendor changes a form, and answer when it gets something wrong. If nobody is named for that, the automation will quietly die and the ROI with it.
The error cost. The one nobody adds. If the automation is wrong one time in fifty and a wrong invoice costs you an hour of cleanup and a frustrated customer, that is a real line on the cost side. Rules-based tasks with a review step keep this small. Tasks where a mistake reaches a customer unreviewed do not.
Run it per task, not per business
The question “what is the ROI of AI for my business” has no useful answer. The question “what is the ROI of automating quote follow-up” does. List the repetitive tasks, run each one through the formula, and sort by return. The top one or two are your first project. Anything below zero stays manual.
This matters more than the arithmetic. The MIT State of AI in Business 2025 study found that only about 5 percent of company AI projects reach daily use. The ones that do are narrow and tied to one job with a measurable result. A per-task ROI list is how you end up on the right side of that number: you pick the narrow project with the clear payback, ship it, and measure it, instead of buying a platform and hoping.
The US Chamber’s small business technology survey put small business AI adoption at 23 percent in 2023, 40 percent in 2024, and 58 percent in 2025, and the Federal Reserve now tracks adoption across the whole economy in its own note. More businesses are trying it every year. The per-task list is what separates trying it from getting paid back.
A worked example
A six-person home services company. The office manager spends about five hours a week chasing quotes and sending follow-ups, three hours on invoicing and payment reminders, and three hours answering the same customer questions by email. Loaded cost on that seat is roughly 35 dollars an hour.
| Task | Hours/week | Yearly cost of the time | Share automated | Yearly value |
|---|---|---|---|---|
| Quote follow-up | 5 | $9,100 | 60% | $5,460 |
| Invoicing and reminders | 3 | $5,460 | 70% | $3,822 |
| Repeat customer questions | 3 | $5,460 | 50% | $2,730 |
Together that is about 12,000 dollars a year of recoverable time. Automating all three in one project at, say, 6,000 dollars to build and 1,200 a year to run gives a first-year return of roughly 4,800 dollars and about 10,800 every year after. Automating only the first one, at perhaps 2,500 to build, pays back in under six months. Either is worth doing. Whether you take the small bite or the bigger one depends on cash and on how much change the office can absorb at once, which is a judgment call, not a formula.
The illustrative costs here are the ranges from our pricing guide. Yours will differ, and the right way to find out is to price the top task properly before committing.
What the number leaves out
The formula counts hours. It does not count:
- Speed. A quote that goes out in an hour instead of two days closes more often. That is revenue, not saved time, and it is often the bigger number.
- Things that were not getting done. Follow-ups that nobody sent because nobody had time are zero hours in the calculator and real money in the business.
- Owner attention. Getting six hours a week back is not six hours of wage. It is six hours of whatever you would do with them.
- Risk. Which data the automation touches, who reviews output, what happens when it is wrong. The NIST AI Risk Management Framework is the long version. The short version is a one-page AI policy that says which tools, which data, and who checks.
Leave these out of the arithmetic, because guessing at them turns a rough number into a made-up one. But hold them in mind when the formula comes out close to zero. A task with a marginal time return and a big speed return is often still worth doing.
When the calculator says no
A negative or marginal return is a useful answer. It usually means one of three things: the task is too small to carry a build cost, it is too irregular to automate reliably, or too much of it is judgment. In the first case, an off-the-shelf tool or a template may get you most of the value for nothing. In the second, wait until the volume is there. In the third, keep the person and give them a better tool, which is the territory of just using ChatGPT.
If what you were about to do was hire for the task, the calculator also answers that. The comparison between a flat software cost and a per-hour human cost is in AI employee vs virtual assistant, and most roles split: software for the repetitive slice, a person for the rest.
Where KGetsIt fits
The calculator is the five-minute version and it is free. The assessment is the same exercise done properly: we go through your actual tasks and books, put real hours and real costs against each one, scope what the top project would cost to build, and hand you a written list sorted by return. It is fixed price, 500 to 3,000 dollars depending on the size of the business, and the fee comes off any build. If the list comes out empty, we say so, and you have spent very little finding out. If you would rather your own team learned to do this, the training covers it, and Massachusetts companies can often have it reimbursed.
Questions people ask
What is the formula for AI ROI?
What is a realistic share of a task that AI can automate?
Should I use the wage or the loaded cost per hour?
How accurate is an online AI ROI calculator?
What if the ROI calculation comes out negative?
Sources
Want this figured out for your business?
The assessment tells you where AI is worth it for you, fixed price, and the fee comes off the build.
Get an assessment →