How to measure the ROI of automation and AI (without guessing)

A simple 4-step way to measure whether an automation or AI tool really paid off: a baseline, the true cost of an hour, every cost, and a check at 90 days.

Ali Hassan

· 6 min read

Quick answer

Measure the task for two weeks before you build: how often it happens, how long it takes and what goes wrong. Price an hour at roughly wage × 1.4 to include employment costs, count every cost including time spent checking AI output, then measure the same way at 30 and 90 days. Monthly saving minus running costs, divided into the build cost, gives your payback time.

Six months after an automation goes live, I like to ask the owner one question: "Did it pay off?"

The most common answer is "I think so. It feels faster." That's not a bad sign, but it's not an answer either. And with AI tools in particular, "it feels faster" can be badly wrong.

Here's a simple way to measure the return on automation or AI, using numbers you already have, so you know whether to keep going, change course or stop.

Why "it feels faster" isn't enough

In 2025, the research group METR ran a careful experiment with experienced software developers working on their own projects. Before starting, the developers expected AI tools to make them about 24% faster. Afterwards, they believed AI had sped them up by about 20%. The measured result? With AI, they took 19% longer.

That's one study, of one kind of work, and other studies find real gains. But it shows something important: people are bad at judging their own time savings. If you want to know, you have to measure.

The good news is that the measuring doesn't need special software. It needs a "before" number and an "after" number.

  1. Measure before

    Time per task, how often it happens, errors and delays. For two weeks, before you build anything.

  2. Price an hour properly

    Wages plus the extra costs of employing someone.

  3. Count every cost

    Build, tools, AI usage, checking time and upkeep.

  4. Measure after

    The same numbers, the same way, at 30 and 90 days.

Four steps to a real ROI number. The most skipped one is the first: without a before, there's nothing to compare.

Step 1: Measure before you build

Pick the task and track it for two weeks. Keep it simple, a tally on a sheet is fine:

  • How many times it happened (invoices sent, leads entered, emails answered).
  • How long each one took, roughly. Time five or ten of them and take the average.
  • What went wrong: typos, missed follow-ups, late replies.
  • How long things waited: how long a lead waited for a reply, how long an invoice sat before being sent.

If you've read the 15-minute test, you've already done a rough version of this. Now you're making it precise enough to check later.

Step 2: Put a real price on an hour

Most people use the hourly wage. That undercounts. Employing someone costs more than their pay.

In the US, the Bureau of Labor Statistics found that private employers spent an average of $46.89 per hour worked on total compensation in June 2026, of which $32.82 was wages and 30% was benefits. So a rough rule is: true cost of an hour ≈ hourly wage × 1.4. Your own number will differ by country and by role, so use your payroll figures if you have them.

Step 3: Count every cost, not just the build

This is where ROI numbers get flattering. Include all of these:

CostOne-off or monthlyEasy to forget?
Building itOne-offNo
Tool subscriptions (Zapier, Make and so on)MonthlySometimes
AI usage (tokens)MonthlyYes, but usually small
Time spent checking AI outputMonthlyVery often
Fixes and changes when apps updateOccasionalYes

The checking time matters most for AI. If AI drafts replies and someone reads every one before sending, that reading time comes off your savings. It's usually still a big win, but count it. For the build and running costs, see what automation really costs and what AI really costs.

Step 4: Measure after, the same way

At 30 days, measure the same things in the same way. At 90 days, do it again. The 30-day number tells you if it works. The 90-day number tells you if it stuck, because some savings fade when people drift back to old habits.

Then the maths is simple:

  1. Monthly saving = hours saved per month × true hourly cost.
  2. Months to pay back = build cost ÷ (monthly saving − monthly running costs).
  3. First-year ROI = (12 months of savings − all first-year costs) ÷ all first-year costs.

Not everything shows up as hours

Time saved is the easiest number, but often not the most valuable one. Track these too, where they apply:

  • Speed: how fast leads get a reply, how fast invoices go out and get paid.
  • Errors: wrong amounts, missed follow-ups, double entries.
  • Money in: more leads converted, fewer late payments.
  • Capacity: could you take on more work without hiring?

Research backs up where the gains tend to show. Stanford's 2026 AI Index notes that productivity gains from AI are largest in structured, measurable work where outputs are easy to monitor. That's exactly the kind of work you can measure before and after.

Measure per person, not just in total

One more thing worth knowing. A large study of customer support staff, published by the NBER, found that an AI assistant raised the number of issues resolved per hour by 14% on average, and by 34% for newer, less experienced staff, with minimal impact on the most experienced people.

So if your total looks small, look closer. The tool might be a big help to your newest hire and do almost nothing for your best person. That's still useful to know: it might change who uses it, or how you train people.

What to do with the result

  • Pays back in under a year: keep it, and look for the next task like it.
  • Pays back slowly but saves errors or speeds up cash: usually still worth keeping. Note the reason.
  • No real saving after 90 days: find out why before adding more. Often the process wasn't fixed first, or people aren't using it. Why automation projects fail covers the usual causes.

Want help setting up the before-and-after numbers for a task you're thinking of automating? Book a free 20-minute call. I'll help you work out what to measure, and give you an honest estimate of the payback.

Share

LinkedInXWhatsApp

Common questions.

  • Work out the monthly saving (hours saved × true hourly cost), subtract the monthly running costs, then divide the build cost by what's left. Under a year is usually a good deal.

Keep readingAll posts