AI ROI: Measure It With a Simple Time Log

To measure AI ROI, time the same task twice: once the way you do it today, and once with AI, counting the minutes you spend checking and fixing the result. The difference in minutes, multiplied by what an hour costs you, minus what the tool costs, is your return. A one-week log is enough to start.

Most small businesses skip this step. Someone says "AI saves us hours," everyone nods, and three months later nobody can say whether the subscription was worth it. This post gives you a simple before-and-after log that works for an insurance agency and for an engineering or architecture firm, with no software beyond a spreadsheet.

Why "we save hours" is not a number yet

AI is already common in independent agencies. A Liberty Mutual study of nearly 1,200 independent agency principals and staff, reported by IA Magazine in August 2026, found that 65% of agents used AI at work in the past year and 41% used it weekly. Users estimated they save about 4 hours a week on average (IA Magazine, Liberty Mutual 2026 Independent Agency Growth Study, national data).

Read that last number carefully. It is an estimate, reported by the people using the tools, and it covers agents across the country. It is a useful signal that time savings are plausible. It is not a measurement of your agency, your tasks or your review time. Liberty Mutual is a carrier, and the article does not state who sponsored the research, so treat the figures as directional.

The same study found that only 18% of agents say their agency has a well-defined AI use policy. Measuring and governing go together: if you cannot say what people use AI for, you cannot measure what it does.

The one-week time log

You need one task, one person (or two), and a spreadsheet with these columns:

For an A&E firm the same table works with a different task: a first draft of a proposal cover letter, a meeting summary, or a summary of an RFQ's requirements. The task matters less than the discipline of timing it the same way twice.

How to run it, step by step

1. Pick one repeatable task. Choose something that happens at least a few times a week, takes 15 to 60 minutes, and has a clear "done." Avoid anything unique each time.
2. Record a baseline first. Time the task five times the way you do it today. Write down the start and end, not a guess afterwards. Use the average.
3. Run the same task with AI. Same type of task, same person if possible. Time two things separately: producing the first draft, and reviewing and fixing it.
4. Count the problems. Note every error you catch: a wrong date, a wrong name, an invented detail. Errors tell you whether the time saved is real or just moved into rework.
5. Do the math. Net minutes saved per task equals baseline minutes minus (AI production minutes plus review minutes).

The formula

Hours saved per month = net minutes saved per task × tasks per month ÷ 60

Value of that time = hours saved × your loaded hourly cost (wages plus benefits and overhead; use a number your bookkeeper agrees with)

ROI = (value of time − monthly cost of the tool) ÷ monthly cost of the tool

A worked example (invented numbers, to show the math)

These figures are made up for illustration. They are not a result from any company.

• Baseline: 30 minutes per task.
• With AI: 8 minutes to produce, 7 minutes to review and fix. Total 15 minutes.
• Net saved: 15 minutes per task.
• Volume: 40 tasks a month, so 600 minutes, or 10 hours.
• Loaded hourly cost: $40, so the time is worth $400.
• Tool cost: $60 a month.
• ROI: ($400 − $60) ÷ $60, about 5.7 to 1.

Now change one input: if review took 20 minutes instead of 7, the net saving drops to 2 minutes, and the ROI nearly disappears. That is why review time goes in the log.

Four traps that make the number lie

• Leaving out review time. A draft that takes 3 minutes and 25 minutes of fixing is not a 3-minute task.
• Cherry-picking easy tasks. If you time only the cases where AI worked well, the average is fiction. Log every attempt, including the failures.
• Ignoring quality. Faster with more errors is not a saving. Keep the "problems found" column honest.
• Counting time that goes nowhere. Ten saved hours only matter if they go into something: more renewals reviewed, more proposals sent, less overtime. Write down what the time was used for.

Keep client data out of the log

The log needs task names and minutes, not client names, policy numbers or project details. And before your team tests anything, make sure everyone knows what must never be pasted into an AI tool.

Where to start this week

Pick the one task your team repeats most and times worst. Today, record the baseline for that task on whatever you do next. By Friday you have five baseline timings; next week you run the same task with AI. Fifteen minutes of setup gives you the first real number your business has had on this.