How do I measure the ROI of AI automation?

Start with a baseline before you automate anything

The most common reason owners can't tell whether AI paid off is that they never wrote down what things looked like before. Two months later, nobody remembers how long the Monday report used to take.

Before you automate a task, spend one or two weeks recording a few numbers. A notepad or simple spreadsheet works.

• Hours per task. How long does the task take each week, and who does it? Ask the person who does it, then check against a real week. People usually underestimate repetitive work.
• Response times. How long does it take to reply to a new lead, a quote request or a customer email? Check the timestamps on your last 20.
• Error rates. How often does something need redoing: a wrong invoice, a missed appointment, a duplicate CRM entry?
• Lead conversion. Out of every 10 inquiries, how many become paying jobs?
• No-shows and late payments. How many appointments get missed each month? How many invoices are past 30 days?

Pick the three or four numbers that relate to what you're automating. If you're automating lead follow-up, response time and conversion matter. If you're automating invoicing, focus on hours and days to get paid.

The ROI formula, in plain English

ROI comes down to three pieces:

1. Time saved. Hours saved per week, times weeks per year, times the loaded hourly cost of whoever used to do the work. Loaded cost means the wage plus what you pay on top of it: payroll taxes, benefits, workers' comp, paid time off. Your bookkeeper can give you the real number for your business.

Two rules here. First, count net hours: if a task used to take five hours and now takes one hour of review, you saved four, not five. Second, saved hours only count if they go somewhere useful. If the time just disappears, the savings are on paper only.

2. Revenue effects. Some automations bring money in, not just save time. Faster replies to leads, reminder texts that cut no-shows, follow-up on unpaid invoices that used to slip. Count these separately, and only when you can trace them.

3. All costs. Software subscriptions, AI credits, any outside help, setup time, and your own time spent managing and reviewing. That last one is the cost people forget most.

Then: (time saved + revenue effects) minus all costs equals net return. Divide net return by all costs to get ROI as a percentage.

What is manual work costing you?Four numbers. See the yearly cost of busywork next to a hire, an agency and a fractional AI manager.
Open the calculator

A worked example

Example: picture an eight-person plumbing company. The office manager earns $28 an hour, and with taxes and benefits her loaded cost works out to about $35 an hour. The owner values his own time at $50 an hour (a modest number on purpose). Before starting, they spent two weeks recording a baseline. Here's a year of results, using 50 working weeks.

Notice the revenue lines use gross profit, not the full invoice amount. A $900 job that costs $550 in parts and labor adds $350, not $900. That's the number that matters.

Now the conservative version: drop every revenue line and count only the hours. That's $18,000 in benefits minus $10,200 in costs, or $7,800 net. Still positive. That's the test I like. If the math only works when the optimistic revenue guesses are included, be skeptical.

These are made-up numbers for one hypothetical business. Your hours, wages and costs will differ. The software line in particular varies a lot by tool, so treat it as a placeholder (as of late 2026, check current pricing). To run your own version in about a minute, try the AI cost calculator. My own pricing is on the pricing page if you want to plug in a real number for outside help.

Leading and lagging indicators

ROI is a lagging number. It takes a few months of data before it means much. In the meantime, you need leading indicators that tell you whether things are on track.

Leading indicators (check weekly): - Tasks completed by agents this week - Hours your team reports getting back - Errors caught in review, and whether that number is falling - Lead response time - How many tasks are still waiting on you (access, decisions, answers)

Lagging indicators (check monthly or quarterly): - Lead conversion rate - Booked jobs or revenue per month - No-show rate - Days to get paid, and invoices past 30 days - Overtime hours, or a hire you didn't need to make yet

If leading indicators look good for a month or two, the lagging ones usually follow. If leading indicators are flat (agents aren't finishing tasks, errors aren't dropping), fix that before you worry about ROI. A simple weekly report that pulls these numbers together helps. My answer on AI reports and dashboards covers how to set one up.

How to count conservatively

Overclaiming ROI is easy, and it leads to bad decisions later. Beyond net hours and gross profit, a few rules keep it honest:

• Only count what you can trace. If leads went up the same month you started a new ad campaign, don't credit all of it to AI.
• Include your own time. If you spend two hours a week managing agents, that's a cost.
• Don't count cash flow as profit. Getting paid faster helps, but money you were going to collect anyway isn't new money. Only count invoices that would have been written off.
• Recheck after a few months. Early estimates are guesses. Replace them with real numbers as soon as you have them.

What this looks like in a real small business

Example: picture a five-person insurance agency that automates three things: quote-request follow-up, renewal reminder emails, and the weekly pipeline report.

Before starting, the agency owner pulls the timestamps on her last 20 quote requests and finds the average first reply takes most of a business day. She records that the pipeline report takes her about two hours every Monday, and that her account manager spends roughly four hours a week on renewal emails.

In the first month, she tracks only leading indicators. The report now arrives Monday morning and takes 15 minutes to review. Renewal emails go out on schedule and the account manager spends about an hour a week checking them. Quote replies go out within minutes during business hours.

By month three, she has enough data to compare conversion rates on quote requests. She counts only the hours in her ROI at first, because one quarter of conversion data isn't much. The hours alone cover the cost. She treats any conversion lift as a bonus until she has six months of numbers to back it up.

The honest part

Some automations don't pay off. A task that takes 20 minutes a month isn't worth automating, no matter how annoying it is. And if a process is broken, automating it just produces broken results faster.

ROI also isn't the only reason to automate. Happier customers and fewer late nights are real but hard to price, so leave them out of the math and mention them separately.

Be careful with vendors (including me) quoting big ROI numbers. Ask to see the math. If the hours saved aren't net of review, or the revenue isn't traceable, the number is marketing.