Estimating ROI before you build an AI automation

Estimate the return on an AI automation in twenty minutes on one page: count the hours the task takes per month, multiply by what an hour costs you, add only the revenue you can attribute to the task conservatively, then subtract the build quote and the monthly running costs. Divide to get months to pay back. Under six months is a strong case for a two-week pilot; six to twelve is fine; over twelve means the scope is wrong, not necessarily the idea. The pilot then replaces the estimate with a measured number.

Why estimate at all?

Because "AI will save us time" isn't a decision. A one-page estimate tells you which task to pilot first, what a reasonable quote looks like, and what number the pilot must hit. It also protects you from the two common mistakes: paying for an automation that saves two hours a month, and refusing one that would pay back in six weeks.

Step 1: hours per month

Pick one task, as in the "what to automate first" guide. Have the person who does it count for one week: minutes per occurrence, occurrences per day. Multiply by four. Include the hidden parts: looking things up, chasing, fixing mistakes. Don't include time the automation won't touch, such as the judgement calls a person will still make.

Step 2: what an hour costs

For an employee, salary plus employer costs, divided by working hours; the Central Bureau of Statistics publishes wage data if you need a benchmark. For your own hours, use what an hour of your time is worth when spent on the business instead. Be honest in both directions; overstating makes every project look good.

Step 3: revenue the task touches

Only for tasks that directly touch sales or attendance: lead replies, quote follow-ups, reminders. Estimate conservatively: "we lose about five leads a month to slow replies; a third would have booked; an average booking is X." If you can't fill a sentence like that, leave revenue at zero and decide on hours alone.

Step 4: costs

Build cost from two or three quotes, as in the cost guide. Running costs from the vendors' pricing pages and the builder's volume estimate: the workflow platform, the messaging provider, AI usage, and a support plan if you take one. Put the support plan in; a build with nobody watching it is not the thing you're valuing.

Step 5: months to pay back

Monthly benefit is hours × cost plus attributed revenue, minus running costs. Payback is build cost divided by monthly benefit. Write it down with the date and the assumptions; you'll compare it against the pilot's measurement.

A worked example (illustrative, not a client)

Change the hours to eight and the revenue to zero, and payback stretches past two years; the same automation, a different business. The numbers are made up to show the arithmetic; yours will differ.

What the estimate can't tell you

Whether it will actually work on your data, whether your team will use it, and whether the exceptions are rare or constant. Those are what the pilot answers. Treat the estimate as the reason to run the pilot, and the pilot's number as the reason to continue.

Best fit and not a good fit

Best fit: owners choosing between two or three tasks, or deciding whether a quote is worth it. Not a good fit: projects whose value is mostly risk reduction or customer experience; estimate those on hours, and accept that part of the return is unmeasured.

What to do next

Fill the five lines for your top task. If payback is under a year, take the page to two or three builders on the marketplace and ask for a two-week pilot that measures the number you used.