How to measure AI ROI: hands-on time vs agent time | TimeUno

How do you measure AI ROI?

Measure AI ROI by comparing what a task cost before AI with what it costs now. Track two numbers for each task: your hands-on time, and the time your AI agents run on their own. Then add the tool cost. The gain is the hands-on time you got back, minus what the tools cost.

The AI ROI formula

AI ROI is the value of what you gained, minus what the AI cost, divided by what the AI cost.

AI ROI (%) = (value gained - AI cost) / AI cost x 100

Value gained is hands-on hours saved times your hourly rate, plus any extra output you shipped. AI cost is the tool bills, the time spent setting up and prompting, and the time spent reviewing and fixing output. Most teams forget the last two.

Step 1: Set a baseline before AI

Pick one repeatable task, like release notes, support replies or weekly reports. Record how long it takes you without AI for two or three weeks. Without a baseline, every "AI saved us time" claim is a guess.

If AI is already in use, estimate the baseline from old timesheets or tickets for the same kind of task, and say it is an estimate.

Step 2: Track hands-on time and agent time

Hands-on time is the time you spend on the work yourself: writing the prompt, reading the result, correcting it, shipping it. This is the cost that lands on your calendar.

Agent time is the time the AI works while you do something else, or wait. It tells you how much work you handed off, and it links to your token and tool costs.

Good AI ROI looks like this: hands-on time per task goes down, agent time per task goes up or holds, and the output stays the same or improves. If hands-on time stays flat, the AI is not saving you time, however busy the agents are.

Why "hours saved" is the wrong number

"AI saved us 10 hours" is a guess, not a measurement. Nobody logged the old hours, and nobody logs the new ones. Self-reported savings also skip the time spent writing prompts, checking output and fixing mistakes.

Agent hours are not your hours either. An agent can run for two hours while you do something else. Count them as yours and AI looks free. Ignore them and you miss what the work costs in tokens and tools.

A worked example

These numbers are illustrative, not a customer result. Say a developer writes weekly release notes and test cases for one product.

The hands-on time dropped by 6 hours. At $60 an hour that is $360 of time back. Minus $40 of tools, the gain is $320 a week.

ROI = ($360 - $40) / $40 x 100 = 800%.

The ratio of agent time to hands-on time shows how much you delegate. Here it is 1.5 to 1. Watch it over months: if it rises while hands-on time falls, you are getting better at using AI.

How to measure AI productivity without timers

Manual timers fail for AI work. You would have to start one for yourself and one for every agent, in every tool. Nobody keeps that up for a week, let alone a quarter.

Record it automatically instead. A time tracker that sees which app is in front of you, and which AI tools are working or waiting, gives you both numbers with no input. TimeUno's AI time tracker does this for tools like Claude Code, Cursor and Codex CLI. It sends metadata only: session times, model, tokens and cost. Prompts, answers and code stay on your device.

What a real week looks like

Here is one real week from a TimeUno account, October 26 to November 1, 2026.

AI time is longer than the week, which is 168 hours. That is not an error. Agents run in parallel, so five agents working for one hour add up to five hours of agent time. It measures how much work was in flight, not how long you waited.

What to log every week

Keep it to four lines, and compare each week with the one before:

1. Hands-on time on the task or project.
2. Agent time on the same project.
3. Tool cost for the week.
4. Output: what shipped, in your own unit (tickets, pages, reports).

Review the trend monthly, not daily. One week can be noise. Three months of hands-on time per shipped unit tells you whether AI pays for itself. To plan what to do with the hours you get back, read time management that works.

Beyond time: what else to count

Time is the easiest return to measure, but not the only one. Add these when they matter to your work:

• Quality: fewer bugs, fewer rewrites, fewer client corrections.
• Speed to ship: days from start to done, not just hours of effort.
• Capacity: work you now take on that you would have turned down.

Measure each against the same baseline. If you cannot measure one, leave it out of the ROI number and note it next to it.

How to report AI ROI to your team or boss

Show one table: hands-on time, agent time, tool cost and output, this month against the baseline. Add the ROI percentage and one sentence on what changed. State your assumptions, like the hourly rate. A short, honest table gets more trust than a big claim.

Common mistakes

• Counting agent hours as saved hours. Agent time is load, not savings.
• Ignoring review time. Checking and fixing output is hands-on time. Count it.
• Measuring once. A single before-and-after hides the learning curve. People get faster at using AI over time, and only a long series shows it.
• Comparing different work. Compare the same kind of task before and after, or the ratio means nothing.

FAQ

What is a good AI ROI?

There is no single number. A positive ROI means the time and output you gained are worth more than the tools and the time spent using them. Track your own trend: if hands-on time per task keeps falling, it is working.

How long before AI shows a return?

Plan for a few months. Early weeks often include setup and learning time, so hands-on time can rise first. Compare month over month, not week over week.

Is agent time a cost or a saving?

Neither on its own. It is a measure of work handed off. The cost shows up in tool and token bills, and the saving shows up in your hands-on time.

Can I measure AI ROI for a team?

Yes. Use the same four lines per project, and add them up. Compare against the baseline for each kind of task, not one team-wide average.

Measure it on your own work

You do not need a spreadsheet to start. See how TimeUno tracks your hands-on time and your AI agent time, then check your own week.