How Do I Measure the ROI of AI and Automation?

How do you measure the ROI of AI and automation? Take what a task costs you today in hours and mistakes, subtract what the automated version costs in software plus the time you spend checking it, and divide that difference by what you spent to build it. The honest answer for most small businesses is that they cannot measure it, because nobody wrote down what the task cost before it got automated.

That is the whole problem in one sentence. This guide gives you the baseline worksheet, the formula, a worked example for a Walworth County contractor, and the three ways this math quietly lies to you.

What Does ROI Mean When You Are Talking About Automation?

ROI, or return on investment, is the value you got back divided by what you spent, written as a percentage. For automation, that value is almost always one of four things: hours saved, revenue recovered, errors avoided, or response time cut. Pick which one you are buying before you buy anything, because a system that saves hours and a system that recovers revenue get measured in completely different ways.

Two terms you will need. Payback period is how many months of savings it takes to cover what you spent to build the thing. Fully loaded hourly cost is what an employee actually costs per hour once you add payroll taxes, insurance, and paid time off, which is typically 25 to 40 percent above the wage itself. Using the raw wage makes every automation look worse than it is.

Here is why this matters more than the technology conversation. McKinsey's State of AI global survey, which polled 1,993 participants across 105 nations in mid-2025, found that 88 percent of organizations regularly use AI in at least one business function, but only 39 percent could attribute any EBIT impact to it, and most of those said the impact was under 5 percent of earnings. Adoption is easy. Proving value is the hard part, and the businesses that prove it are the ones that measured before they started.

The four value types, with what each one actually looks like:

• Hours saved. A person stops doing a repetitive task. Real only if those hours get redeployed or you stop paying for them.
• Revenue recovered. Quotes that used to go cold now get followed up. Bookings that used to get missed at 9 p.m. now get captured. This is the highest-value category and the easiest to prove.
• Errors avoided. Double bookings, wrong addresses on invoices, missed permit deadlines. Price each error at what it cost you the last three times it happened.
• Response time cut. Answering a lead in five minutes instead of five hours. Measure this as a conversion-rate change, not as a feeling.

What Numbers Do I Need Before I Automate Anything?

You need five numbers per task, and you need them written down with a date. Without a dated baseline you are guessing forever, and guessing is how a $200 a month tool survives three years without ever earning its keep.

The Five-Number Baseline

1. Volume. How many times does this happen in a month? Count it, do not estimate it. Quote requests, invoices sent, appointment reminders, review replies.
2. Minutes each. Time yourself doing it five times and take the median, not the average. One horrible outlier will skew the average and make your ROI look fake later.
3. Who does it. The owner, a $22 an hour front-desk person, and a $65 an hour technician are three different economics.
4. Fully loaded hourly cost. Wage times 1.25 to 1.4. For owner time, use the hourly rate of the work you would do instead, which is usually selling or producing, not the wage you pay yourself.
5. Leak rate. How often does this task get skipped, done late, or done wrong, and what did that cost? If you send 40 quotes a month and 12 never get followed up, that is your number.

How to Collect the Baseline in Two Weeks

Put a single shared note or spreadsheet where the work happens. Every time somebody does the task, they add one line: date, how long it took, and whether anything went wrong. Two weeks is enough for anything that happens daily. For monthly tasks like invoicing or reporting, use last month's calendar and your own honest reconstruction, then confirm it next cycle.

Our post on which tasks to automate first covers how to pick which processes deserve this effort. If you baseline everything, you will baseline nothing.

How Do I Calculate the Return on a Single Automation?

Use two numbers: annual net benefit and payback period. Annual net benefit is your yearly savings minus your yearly cost. Payback period is the one-time build cost divided by the monthly net saving. If payback is under six months, build it. If it is over eighteen, do not.

The Formula

1. Annual gross benefit equals (monthly volume times 12) times (minutes saved per item divided by 60) times (fully loaded hourly cost), plus any recovered revenue or avoided-error cost.
2. Annual running cost equals software subscriptions plus the AI usage you pay for plus the hours per month somebody spends reviewing, fixing, and maintaining the automation, priced at their fully loaded rate.
3. Annual net benefit equals gross benefit minus running cost.
4. Payback months equals one-time build cost divided by (annual net benefit divided by 12).
5. ROI percent equals annual net benefit divided by (annual running cost plus build cost) times 100.

The step everyone skips is step two's last clause. Every AI automation has a review tax: the minutes a human spends reading what the machine produced before it goes out. Early on that tax is high, sometimes half the time you saved. It falls as you tune prompts and rules, but it never hits zero, and a projection that assumes zero is a projection you will regret in a budget meeting.

What Each Piece Costs

As of February 2026, check current pricing: a general AI assistant on a business plan runs roughly $20 to $30 per user per month (ChatGPT Business and Claude Team are both in that range, and Gemini comes bundled into Google Workspace Business plans, which start around $14 per user per month on annual billing). An automation platform such as Zapier starts around $20 a month billed annually for a small task allowance, and Make's paid tiers start around $12 a month. Build cost is either your own hours or an outside builder, which for a small business automation typically lands somewhere between a few hundred and a few thousand dollars depending on how many systems it has to touch. Our guide to what AI actually costs a small business breaks the line items down further.

What Counts as a Real Saving, and What Is Fake?

A saving is real when it changes a bank statement or a sales number. Hours saved are only money when those hours get spent on something that earns, or when you genuinely stop paying for them. Everything else is a feeling, and feelings do not survive a bad quarter.

Run every claimed saving through three tests:

• The payroll test. Did anyone's hours, headcount, or overtime actually change? If a part-timer went from 25 hours to 18, that is real. If everybody works the same hours and just feels less swamped, that is worth something, but it is not $14,000 a year.
• The redeployment test. If hours were freed but nobody was let go, name the revenue-producing thing those hours went into. "The office manager now makes 20 outbound calls a week" is a real answer. "She is less stressed" is a benefit, but log it as a retention benefit, not as dollars.
• The traceability test. For recovered revenue, can you point at specific closed jobs that came from the automated follow-up? If your system tags them, you can. If it does not tag them, add the tag before you claim the number.

Also count the costs people forget: the hours you spent learning the tool, the month things were broken, the subscription you kept paying for the old way of doing it, and the annual price increase.

What Does This Look Like for a Lake Geneva Business?

Picture a roofing contractor in Elkhorn with an owner, an office manager, and three crews. The complaint is not "we need AI." The complaint is that quotes go out and then vanish.

The baseline, collected over two weeks and written down:

• Volume: 46 quotes a month in season.
• Follow-up: the office manager intends to follow up twice per quote at about 6 minutes each. In practice, roughly 15 of 46 get any follow-up at all.
• Fully loaded cost of the office manager: $24 an hour wage, about $31 loaded.
• Close rate on quotes that got followed up: 31 percent. On quotes that did not: 12 percent.
• Average job value: $9,400.

The automation: when a quote is marked sent in the CRM, a workflow sends a personal-sounding follow-up at day 3, day 8, and day 21. An AI step drafts each message using the roof type, the address, and the quoted price, and the office manager approves the batch every morning in about 10 minutes. Anything that gets a reply drops out of the sequence and goes to her inbox.

The math, one year out:

• Software: automation platform and AI assistant, about $75 a month, so $900 a year.
• Build: about $1,800 paid once, plus 6 hours of the owner's time.
• Review tax: 10 minutes a day, five days a week, at $31 loaded, which is roughly $1,340 a year.
• Time saved on manual follow-up: modest, maybe $700 a year, because she was not doing most of it anyway.
• Revenue effect: the 31 quotes a month that used to go unfollowed now get three touches. If even four extra jobs a year close because of it, that is $37,600 in additional revenue. At a 22 percent gross margin, roughly $8,270 in gross profit.

Net benefit in year one is about $6,730 against a $1,800 build and $2,240 in running cost, and payback lands under five months. Notice which number carried the whole thing. It was not hours saved. It was revenue that used to leak out the side of the process. That is the pattern in almost every small business automation worth building.

Also notice what would have made this unmeasurable: if the CRM did not tag which jobs came from the follow-up sequence, the owner would be left saying "it feels like we are closing more," which is exactly the position most businesses end up in.

How Long Should I Wait Before Judging an Automation?

Give it 30 days to stop breaking, 90 days to produce a trustworthy number, and 180 days to decide whether it stays. Judging at week two is how good automations get killed and bad ones get praised.

1. Day 30 check: does it run? Look at error rates, not results. How many times did it fail, fire twice, or send something embarrassing? Fix those before you look at outcomes.
2. Day 90 check: is the review tax falling? Compare the minutes-of-review log from month one to month three. Flat or rising review time means the design is wrong.
3. Day 180 check: does the money show up? Now compare against your dated baseline. If the net benefit is negative and you have already tuned it twice, turn it off. Turning things off is a skill.

What Should I Put on a One-Page Monthly Scorecard?

Six lines, same six every month, on one page you can read in two minutes. Anything longer will not get read, and a scorecard nobody reads is a scorecard that does not exist.

• Runs and failures. How many times each automation fired, and how many errored.
• Review minutes. Total human time spent checking or fixing AI output this month.
• Volume handled. Quotes followed up, invoices chased, reviews replied to, messages answered.
• Outcome metric. The one number this automation exists to move: close rate, days to payment, no-show rate, response time.
• Cost. Every subscription that touches this workflow, added up, including the ones you forgot.
• One thing to fix. The single worst behavior you noticed this month.

Build this before you build your second automation. Businesses that stack five automations with no scorecard end up with a monthly software bill they cannot justify and no idea which piece is carrying the load.

Do This This Week

1. Pick one task that annoys you and happens at least 20 times a month. Just one.
2. Start a two-week tally sheet for it: date, minutes, and whether anything went wrong. Put it where the work actually happens.
3. Calculate the fully loaded hourly cost of whoever does that task (wage times 1.25 to 1.4).
4. Write down the leak: how often the task gets skipped or done late, and what that has cost you in the last year.
5. Name the one outcome metric this automation would move, and confirm you can actually pull that number today.
6. Add up every subscription you already pay that touches this workflow, so your "before" cost is honest.
7. Set a calendar reminder for 90 days out titled "check the automation numbers," and put the baseline figures in the invite so you cannot lose them.

Where to Go From Here

The decision is not whether AI is worth it in general. It is whether one specific process in your business costs more than the tool that would replace it, and whether you can prove it six months from now. That comes down to a dated baseline, a formula that includes the review tax, and the discipline to turn off what does not pay.

If you would rather not build the worksheet yourself, our AI Opportunity Audit maps every repetitive process in your business, puts an hours-and-dollars number on each one, and ranks them by payback period so you know what to build first and what to leave alone. And if you are not sure customers can even find you yet, start with the free Local Visibility Audit on our homepage. Fixing a leaky funnel usually beats automating a small one.

Sources and Further Reading

1. The State of AI: Global Survey 2025. McKinsey and Company, 2025.
2. Survey Reveals Small Businesses Are Using AI to Boost Productivity. Intuit QuickBooks, June 2025.
3. Plans and Pricing. Zapier, February 2026.
4. Pricing and Subscription Packages. Make, February 2026.
5. Pricing. OpenAI, February 2026.
6. Plans and Pricing. Anthropic, February 2026.
7. Compare Flexible Pricing Plan Options. Google Workspace, February 2026.