AI automation ROI calculator

Four numbers from you, and the arithmetic an accountant would do on a napkin, written out so you can check every step:

• Hours back a week: hours on the task times the share a machine could do.
• Hours back a year: that, times the weeks a year the task runs. The default is 48, which leaves room for leave and holidays. Change it if your task runs all year or only in season.
• What those hours are worth: hours back a year times what an hour costs you.
• Setup cost: hours to set it up, costed at the same hourly rate, plus any one-off spend.
• Payback: setup cost divided by what you save each week after the tool's own running cost.

Everything is your number. The calculator doesn't assume a saving, a wage or a success rate, and it doesn't round in the tool's favour.

The hours figure is where most estimates go wrong, and they go wrong low. People count the time they spend doing a task and forget the time around it: finding the file, switching between systems, checking the result, chasing the missing piece.

Measure one ordinary week. Everyone who touches the task writes down the minutes each time they do. Add it up. If the task is seasonal, take a busy week and a quiet one and use the average.

Count everyone. A task that takes the office manager two hours and the owner one hour to check is three hours a week, not two.

Use pay plus overheads, not pay alone. Payroll taxes, benefits, software seats and the desk the person sits at all scale with their hours. If you don't know your overhead, adding a quarter to the hourly wage is a reasonable place to start, and you can check it against your books later.

For your own time, use what you would pay someone to do the work, not what you bill clients. An owner's hour is worth more than that, but only if it goes somewhere better. If the hours saved on invoicing go into more invoicing, the saving is the wage, not the billing rate.

This is the number that decides the answer, and the one vendors inflate. Very few tasks go to zero. Someone still reviews the output, handles the exceptions and fixes the run that broke.

A useful way to estimate it: write out every step of the task, mark the steps a machine could do without a judgement call, and count the minutes in those steps. Then take some off for checking. If the task is mostly judgement, the share is low, and the calculator will say it isn't worth automating yet. That's a useful answer too.

The result always shows the same task at half the share you entered. Plan on that number. If the automation still pays back at half, it's a safe bet; if it only works at the full share, you're betting on the vendor's demo.

Setup hours cover mapping the process, building the automation, testing it on real work, fixing what the test found and showing the team how it runs. On a first project, that's usually more hours than the build itself.

Put anything paid once into one-off spend: a contractor, a template, an integration fee. Put the tool's subscription into the monthly cost. The calculator takes the running cost off the saving every week, so a tool that costs more than it saves never pays back, and the result says so.

Our rules of thumb, which you should disagree with if your business says otherwise:

• Under three months, at half the share: do it this quarter.
• Three to twelve months: worth doing if the task is stable and nobody is about to change how it runs.
• Over a year: look for a cheaper way first, often a template, a checklist or a better use of software you already pay for.
• Never: the tool costs more than the hours it saves. Keep the hours, or find a bigger task.

Money isn't the only return. Hours back a year is the number to put next to what you'd do with them. Forty hours a year is a week. Four hundred is a quarter of someone's job, and a different conversation about what that person could do instead.

Hours are the part of the return you can count before you start. Four other parts are real and harder to price:

• Speed. A quote that goes out in ten minutes instead of two days wins work the hours figure never sees.
• Errors. Retyping produces mistakes, and a wrong figure on an invoice or an order costs more than the minutes it took to type.
• Work nobody does now. Some jobs get skipped because they take too long: following up every quote, answering every review. An automation can make them happen at all, which shows up in revenue, not in hours.
• Upkeep. An automation that fails quietly costs more than the task did. Put an hour a month into checking it still runs, and count it in the setup or the running cost.

Keep these as reasons, and let the calculator decide on hours alone. If a project only pays back once you add them, say so out loud before you buy it.

The best first task is frequent, follows the same steps each time, starts from information that is already digital, and does little damage if a run goes wrong. Retyping orders from email into a spreadsheet is a good first task. Deciding which customers get a discount is not.

If you're not sure which of your tasks that is, the AI readiness assessment takes about four minutes and ranks the three things to fix first, starting with the weakest area of how your business runs.