AI automation ROI: how to work it out before you buy
By Saiful Islam Shah. Published . 3 minute read.
AI automation ROI is the value an automation returns compared with what it costs to build and run. For a small business you can estimate it in ten minutes with four numbers: hours saved per month, the cost of an hour, the build fee and the monthly running cost. This guide gives the formula, a worked example and the cases where the answer is "do not automate this".
The AI automation ROI formula
Monthly saving = hours saved per month × cost of one hour
Payback period = build fee ÷ (monthly saving − monthly running cost)
First-year return = (12 × monthly saving) − build fee − (12 × monthly running cost)
Use the full cost of an hour, including salary, taxes and overhead, not only the wage.
A worked example
The numbers below are an illustration, not a client result. Put in your own.
A coordinator spends 10 hours a week copying website enquiries into the CRM and sending a first reply. Her time costs the business $8 an hour.
| Item | Amount |
|---|---|
| Hours saved per month | 10 hours × 4.3 weeks = 43 hours |
| Monthly saving | 43 × $8 = $344 |
| Build fee (one workflow) | $500 |
| Monthly running cost (platform and AI usage) | $40 |
| Payback period | $500 ÷ ($344 − $40) = about 1.6 months |
| First-year return | (12 × $344) − $500 − (12 × $40) = $3,148 |
With a higher hourly cost or a higher build fee the result changes, which is the point of doing the sum. For typical build fees see how much AI automation costs.
Returns that are easy to miss
Hours saved is the simplest measure but often not the largest.
- Faster replies to leads. A lead answered in minutes is more likely to buy than one answered the next day. If automation raises your conversion rate, that gain can outweigh the time saved.
- Fewer errors. A wrong address or a missed invoice has a real cost in refunds, re-shipping and goodwill.
- Work that was never done. Follow-ups, review requests and reminders that staff skipped when busy now happen every time.
- Out-of-hours coverage. Enquiries at night and at weekends get an answer.
- Capacity. The team can handle more customers without another hire.
When you estimate these, be conservative and write down your assumption so you can check it later.
Signs a project will not pay back
- The task is rare. Something done twice a month for ten minutes is not worth building for.
- The process keeps changing. Each change means paying to rebuild.
- Every case is an exception. If no two are handled the same way, there is nothing to automate yet.
- The data is not there. Automation cannot use information that lives in people's heads.
- Nobody will own it. A workflow with no owner breaks quietly and stays broken.
Our guide on what to automate first shows how to rank tasks so you start with the one that pays back fastest.
How to measure after launch
Decide the measures before you build, and record the "before" figure.
- Time: hours spent on the task per week, before and after.
- Speed: average time to first reply, or time from order to dispatch.
- Volume: how many items the workflow handled without a person.
- Errors: how many needed correcting.
- Outcome: conversion rate, tickets per customer or overdue invoices, whichever the project was meant to move.
Review at 30 and 90 days. If the numbers are not moving, find out why before automating the next thing.
A sensible first target
Look for a project that pays for itself within six months on time saved alone. Anything it adds in conversions or fewer errors is then a bonus, and you have a real result to judge the next project by. Our AI automation pricing page lists fixed build fees you can put straight into the formula.