How to calculate automation ROI (with a worked example)
Most automation ROI estimates are either wildly optimistic or never done at all. Here's a simple, honest method — with a worked example you can copy.
On this page
Automation projects are usually justified with one of two numbers: a vague promise (“it'll save loads of time”) or a spreadsheet so optimistic nobody believes it. Neither helps you decide. What you need is a simple, honest calculation you can defend — and that takes about fifteen minutes.
This guide walks through the method we use when we scope projects, then works through a complete example.
01The formula
ROI (return on investment) is the net gain from a project as a percentage of what it cost. Payback period is how long it takes to recover the up-front cost. For automation, both come from the same four ingredients:
- 01Time recovered — hours per year the process takes today, multiplied by the share you'll automate.
- 02Value of that time — hours recovered × a loaded hourly cost.
- 03Errors avoided — how often mistakes happen × what each one costs × the share automated.
- 04Costs — the one-off build, plus ongoing running and maintenance costs.
02Step 1: Measure the task, don't guess it
You need three numbers: how many times the task happens per week, how many minutes it takes each time, and how many people do it. The most common mistake is estimating these from memory. Instead, time it for a week — even a rough tally is far better than a guess.
Include the whole task, not just the obvious part. Processing an invoice isn't only typing it in; it's opening the email, finding the purchase order, checking the amounts, chasing the missing detail and filing it.
03Step 2: Use a loaded hourly cost
A loaded hourly cost is what an hour of someone's time really costs the business: salary plus employer taxes, benefits, equipment, software and a share of management time. A common rule of thumb is 1.2–1.4× the base hourly wage. Use your own figure if you have one — finance teams usually do.
04Step 3: Be realistic about the share automated
Very few processes are 100% automatable. Exceptions, approvals and edge cases will still reach a person. Estimate the share of runs that will go end-to-end without anyone touching them, and use that — not the best case. If you're unsure, 60–80% is a sensible planning range for a well-defined, rules-based process; lower for work that needs judgement.
05Step 4: Put a price on errors
Manual work produces mistakes: a wrong figure, a missed follow-up, a duplicated record. Each one costs something — rework, a refund, a lost customer. If you can estimate how often errors happen and what they cost, include it. If you can't, leave it out and treat your result as conservative.
06Step 5: Count every cost
- Build cost — design, development, testing and rollout.
- Running cost — hosting, software licences, and model usage if AI is involved.
- Maintenance — updates when connected systems change their APIs or your process changes.
- Internal time — the hours your team spends defining the process and testing the result.
07A worked example
Here's a hypothetical but typical case: a finance assistant processing supplier invoices by hand. Every figure below is an assumption you'd replace with your own.
| Input | Value |
|---|---|
| Invoices per week | 60 |
| Minutes per invoice | 8 |
| People involved | 1 |
| Loaded hourly cost | $30 |
| Share automated | 80% |
| Invoices with an error | 2% |
| Cost of each error | $40 |
| Build cost | $8,000 |
| Running cost | $120 / month |
Hours per year: 60 × 52 × 8 ÷ 60 = 416 hours. At 80% automated, that's about 333 hours recovered — roughly eight working weeks.
Value of time: 333 × $30 ≈ $9,984 a year.
Errors avoided: 3,120 invoices × 2% × $40 × 80% ≈ $1,997 a year.
Annual value: about $11,981, or $998 a month. Subtract the $120 running cost and the net gain is about $878 a month.
| Result | Value |
|---|---|
| Annual value | ≈ $11,981 |
| Net gain per month | ≈ $878 |
| Payback period | Month 10 |
| Net value after 3 years | ≈ $23,600 |
| 3-year ROI | ≈ 192% |
Payback: $8,000 ÷ $878 ≈ 9.1 months, so the build has paid for itself during month 10. Over three years the net value is about $23,600 against total costs of $12,320 — an ROI of roughly 192%.
08Payback or ROI: which matters more?
They answer different questions. Payback tells you how long your money is at risk — useful when cash is tight. ROI tells you how much the project returns over its life — useful when comparing projects. A project with a short payback and a modest ROI is often a better first automation than a large one with a big ROI and a long payback, because it proves the approach quickly.
09Common mistakes
- Assuming 100% automation. Exceptions always exist. Plan for them.
- Using base salary instead of loaded cost. It understates the value of time.
- Forgetting running costs. Especially model usage for AI systems, which scales with volume.
- Counting hours the team won't actually get back. Five minutes saved across fifty different tasks is harder to redeploy than one four-hour block.
- Skipping the measurement. An ROI built on a guessed task time is a guess with a percentage sign.
10Your next step
Pick the one process you suspect costs the most, time it for a week, and run the numbers. If you're not sure which process that is, the manual hours audit ranks your options in a couple of minutes. And if the numbers look promising, we can scope it with you — see our workflow automation service.