To measure AI ROI in a small business, pick one job, record how it performs for a few weeks before the AI is switched on, record it again afterwards, and put a dollar figure on the difference using three tests: time saved, money made and money kept. Subtract everything the AI costs to run, including the time it takes someone to look after it. If the figure is not clearly positive after a fair trial, stop or change it.
That sounds obvious, yet most AI spend in small firms is never measured at all. People feel busier or less busy and leave it there. The method below needs nothing more than a spreadsheet and the discipline to take a baseline first.
Step one: choose one job and write down the baseline
Measure a single job, not “AI across the business”. Good candidates have a count and a clock: enquiries answered, quotes sent, invoices chased, hours spent on admin. Before anything changes, record for three or four weeks:
- how many times the job happens;
- how long each one takes, and who does it;
- how often it goes wrong or gets missed;
- what a typical result is worth, such as an average job value or fee.
If you skip this step, you cannot tell afterwards whether anything improved. It is also the cheapest step, and the one people rush.
The three tests
An AI project earns its keep by passing at least one of these. It is worth being strict, because a project that passes none is a hobby.
Test 1: time saved
Time saved only counts when it turns into something. Two hours a week saved on admin is worth the hourly cost of that person only if those hours go to work that earns, or if you would otherwise have paid for extra hours or a new hire. If the saved time simply disappears into a longer lunch, the benefit is real but it is not a financial return.
Example, with invented numbers for illustration: an office manager spends six hours a week typing job notes into quotes. A drafting step cuts it to two, and she now reads and corrects the drafts. If her time costs the business $45 an hour, four hours is $180 a week. Take off what the tool and its upkeep cost per week to see the net figure.
Test 2: money made
This is extra revenue that would not have come in otherwise. The cleanest case is enquiries that used to be missed. Suppose, as an example, a plumbing business misses about five calls a week after hours, and one in five of those callers would have booked a job worth $400. That is $400 a week of work lost. If an after-hours assistant catches the calls, the question is how many of them become jobs, which you can count from the booking records in a month.
Be careful about attributing. If revenue rises in the same month you launch the AI, check for other causes such as season or a marketing push, and use your baseline rather than your impression.
Test 3: money kept
This is money the business already earns that currently leaks. Common leaks are overdue invoices, no-shows, quotes that never get followed up and customers who drift away because nobody contacted them. A reminder system that reduces late payments by a few days improves cash flow. An appointment reminder that cuts no-shows keeps revenue that was already booked. These are often the easiest returns to measure, because the baseline sits in your accounting or booking software. Our guides on reducing no-shows and getting paid faster show the patterns.
Counting what the AI costs
A fair AI ROI calculation counts every cost, not just the subscription:
- the build or set-up charge, spread over a sensible period such as twelve months;
- monthly fees for the platform, usage and hosting;
- the time staff spend reviewing, correcting and updating it;
- the cost of mistakes, such as a refund or a lost customer from a wrong answer;
- any extra software needed to run it.
The review time is the item most often forgotten. If a person reads every AI draft, that is a cost that belongs in the sum. Our guide to what AI automation costs covers the typical shape of the fees.
The AI ROI sum
Add up the value from the three tests over a period, subtract the total cost for the same period, and divide the result by the cost. Using the invented figures from above: if the quote drafting saves $180 a week, and the tool, upkeep and review time cost $60 a week, the net is $120 a week on a $60 weekly cost, a return of two times. Your figures will differ. What matters is that each number comes from a record, not a guess.
Two checks keep the sum fair. First, ask whether the benefit would have happened anyway. Second, ask whether you could get the same result with something cheaper, like a template or a checklist. The related automation ROI calculator guide works through the arithmetic for fixed-step automations.
A one-page tracker
You do not need software for this. A single sheet with one row per week is enough:
| Column | What goes in it |
|---|---|
| Volume | How many times the job ran |
| Time | Staff hours spent on it, including review |
| Result | Bookings, quotes accepted, invoices paid, whatever the job aims at |
| Misses | Enquiries lost, errors found, complaints |
| Cost | Fees and usage for the week |
Fill in the baseline weeks first, then the trial weeks, and put the two side by side. Ask the person closest to the work for their view as well. Staff who use the tool daily often spot a problem, such as a wrong answer that keeps repeating, long before the figures do.
When to stop, change or double down
- Stop if after a fair trial of six to eight weeks no test is passed and you cannot name what would change that.
- Change if one test is passed but the cost is too high. Narrow the job, drop a feature or cut the review load.
- Double down if the figure is clearly positive and the quality checks hold. Move to the next job on the list.
Write the decision date down at the start. Without one, trials drift on for months because nobody wants to be the person who calls it. A date also stops a half-working tool being kept alive by hope.
Do not judge only on usage. A tool that staff use daily can still lose money, and one that runs quietly in the background can pay for itself many times.
Projects that are hard to measure
Some AI uses give no clean number: a general chat tool staff use however they like, or an assistant that makes people feel faster. If you want to keep them, run them as a small, capped expense and review the cost yearly. Do not expand them on the strength of enthusiasm. The broader case for where AI pays and where it does not is in our piece on whether AI is worth it for a small business.
If you would like help picking the first job and setting up the baseline, CentriWeb builds and looks after AI systems for businesses, and the starting point is a 30 minute call about the part of the business that costs you most.