Measuring automation ROI honestly

Every automation vendor has a slide with a big number on it. We’d rather show you how the number was built, because the honest version is usually smaller — and still worth it.

The simple formula

Our calculator uses the same arithmetic you would:

hours per month = tasks per week × weeks per month × minutes per task ÷ 60 × share automated × (1 − review time)

Multiply by the cost of an hour and you get the value per month. Subtract running costs — model usage, hosting, any subscription — and you get the net. Divide the build price by the net and you get the payback period in months.

That’s it. No “productivity multiplier”, no “opportunity cost of delayed revenue”.

The three numbers people leave out

Review time. People still look at what the automation does — approving refunds, checking flagged invoices, reading hand-offs. In the first months that’s real time. We assume 15% of the saved time goes back into review and adjust once we have data.

Running costs. Model calls, hosting, the workflow tool’s subscription, monitoring. For a small team they’re usually a few hundred dollars a month, but on high-volume or document-heavy work they can be the biggest line. Estimate them before you build, not after.

Work that moved, not vanished. If an agent answers customers faster but creates more follow-up questions, the hours reappear somewhere else. Track re-opened conversations and downstream tickets, not just the conversations the agent closed.

Measure before you build

You can’t show a before-and-after without a before. For two weeks before any build, we count:

• how many times the task happens,

• how long it takes (a sample of timed runs beats a guess),

• how often it goes wrong and what fixing that costs.

It’s dull, and it’s the single thing that makes the result believable to a finance team.

Report ranges, not points

“We’ll save 61,000 hours” sounds precise and isn’t. A range — “between 40 and 70 hours a month at current volumes, assuming 15% review” — is more honest and easier to defend when volumes change. Our calculator shows one number for readability, but the assumptions are always printed next to it, and you can change every one.

Look again at three months

The first month flatters nobody: people are learning to trust the system, thresholds are cautious, edge cases are still being found. Measure again at three months with the same method as the baseline. That’s the number to put in a board deck.

A note on the numbers on this site

Every figure on this template is illustrative — invented for a fictional agency so you can see how the layout works with real-looking data. Replace them with your own measured results, keep the “results vary” note next to them, and never promise a client a number you haven’t measured.