Every automation proposal comes with a savings figure. Few come with the working behind it. If you're the one signing off, you should calculate automation ROI yourself, on your own numbers, before anyone writes a line of code. It takes an afternoon and a spreadsheet.
This guide walks through the calculation in six steps, with a worked example, then covers the mistakes that make a weak case look strong.
Why published automation ROI figures vary so much
The headline numbers are all over the place. In a 2016 McKinsey interview, LSE professor Leslie Willcocks reported first-year returns from robotic process automation ranging from 30% to as much as 200% across the case studies he examined. A range that wide tells you the answer depends almost entirely on the process you pick and how honestly you count.
Many companies don't count at all. Deloitte's 2022 intelligent automation survey found that over half of respondents had not calculated cost reduction, and 70% had not calculated the expected increase in revenue. It also found that the average payback period for organizations piloting automation rose from 16 months in 2020 to 22 months in 2021/22. Plan for payback in months or years, and build your case to survive that.
How to calculate automation ROI in six steps
Step 1: Measure the process as it runs today
Pick one process and count, for a typical month:
How many items go through it: invoices, orders, claims, applications.
How long each one takes from start to finish, including the lookups and the chasing.
How many need rework because of an error.
Time a handful of real items with the people who do the work. Managers' estimates usually come in low, because they miss the small interruptions.
Step 2: Put a loaded cost on an hour
Use the full cost of an hour of staff time: salary plus payroll taxes, benefits, equipment and office space. Your finance team will have a figure. Using salary alone makes the savings look smaller than they are.
Step 3: Decide what share the automation will really handle
No automation handles every case. Some items will be unusual and go to a person, and someone still reviews the automated ones. Start with a cautious estimate, then replace it with a measured figure once the automation has been tested on 20 to 50 of your own real examples.
Step 4: Count every cost, including the dull ones
One-off costs | Ongoing costs |
|---|---|
Process mapping and discovery | Hosting and software licenses |
Build and integration with your systems | AI usage fees, if a step uses AI |
Testing on real cases | Maintenance when connected systems or screens change |
Training and changes to how the team works | Time for the person on your side who owns the automation |
Step 5: Add the benefits that aren't hours
Fewer errors and less rework.
Faster cycle times: invoices paid on time, early-payment discounts taken, orders shipped sooner.
Growth without hiring, because the same team handles more volume.
Cleaner records when the auditors arrive.
Count these only where you can attach a figure you'd defend. If you can't, list them as upside and keep them out of the core sum.
Step 6: Run the two numbers that matter
Three-year ROI = (net benefit over three years − one-off cost) ÷ one-off cost.
Payback period = one-off cost ÷ monthly net benefit.
Net benefit means savings minus ongoing costs. Three years is a fair window, because most automations take a few months to settle and the build cost needs time to earn back.
A worked example
The figures below are made up to show the method. They are not a quote and don't describe any real company.
An accounts payable team processes 1,500 supplier invoices a month. Each takes about 6 minutes to check against the purchase order and delivery note, then enter. That's 150 hours a month.
After automation, 80% of invoices match automatically and need a one-minute review. The other 20% still take the full 6 minutes.
Before | After | |
|---|---|---|
Invoices needing full handling | 1,500 | 300 |
Hours on full handling | 150 | 30 |
Hours on quick reviews | 0 | 20 |
Total hours a month | 150 | 50 |
That frees 100 hours a month, or 1,200 hours a year. At a loaded cost of $40 an hour, it's worth $48,000 a year.
Now the costs. Say the build costs $60,000 and running costs (hosting, licenses and upkeep) come to $12,000 a year. Net benefit is $36,000 a year, or $3,000 a month.
Payback: $60,000 ÷ $3,000 = 20 months.
Three-year ROI: ($108,000 − $60,000) ÷ $60,000 = 80%.
Test how fragile the case is
Before you trust the result, rerun it with a lower and a higher automated share. Everything else stays the same.
Share matched automatically | Hours saved a month | Annual saving | Payback | Three-year ROI |
|---|---|---|---|---|
60% | 75 | $36,000 | 30 months | 20% |
80% | 100 | $48,000 | 20 months | 80% |
90% | 112.5 | $54,000 | About 17 months | 110% |
That's a solid, unspectacular case, and the kind you can defend to a board. If the team now captures early-payment discounts it used to miss, the number rises. If only 60% of invoices match automatically, the case gets thinner, which is why Step 3 deserves the most care.
Mistakes that make a weak case look strong
Counting hours that won't actually be freed. Saving ten minutes a day for each of 20 people rarely turns into real capacity. Saving one person half their week does.
Leaving out upkeep. Systems update, portals get redesigned and rules change. Somebody pays for that every year.
Assuming everything gets automated. Exceptions are where the hardest work sits, and they don't disappear.
Automating a broken process. Deloitte's survey named process fragmentation as the top barrier to scaling. Tidy the process first, or the automation repeats the mess faster.
Ignoring your people. If the team doesn't trust the output, they'll recheck everything by hand and the savings vanish. Show them the log of what the automation did and let them approve the exceptions.
Which processes tend to give the strongest automation ROI
The room for automation is real. McKinsey's 2017 study found that almost half of work activities could be automated by adapting technologies that had already been demonstrated, and its 2023 analysis estimated that generative AI and other current technologies could automate activities that take up 60% to 70% of employees' time. Potential is still a long way from payback. The processes that usually earn their cost back sooner share these traits:
High volume, every week rather than once a quarter.
Clear rules your team can write down.
Inputs that arrive digitally, even as messy PDFs or emails.
Expensive errors, such as duplicate payments or missed deadlines.
Several systems involved, with someone copying data between them.
Month-end close, invoice matching, order entry, claims handling and expense approvals are common candidates. Our workflow automation page shows how these run once automated, with people approving only the exceptions. Where the only way into a system is its screens, RPA may suit that step better.
How to present the case to your board or partners
Keep it to one page. Show the process as it runs today, with the hours and error rates you measured. Show the cost table from Step 4 and the sensitivity table, so nobody has to take the middle figure on trust.
Name the person who will own the automation, and say what you'll measure in the first three months to prove it's working. A short case with honest assumptions gets approved faster than a long one with a heroic number.
A checklist before you sign
Have you timed the process with the people who actually do it?
Is the hourly cost fully loaded?
Is the automated share a guess, or tested on your own examples?
Are running and maintenance costs in the sum?
Does the payback still work if savings come in 30% lower than planned?
Do you know who on your team will own the automation after launch?
If you'd rather not build the spreadsheet from scratch, our free 5-day audit looks at one department, ranks its tasks by hours saved and ends with a fixed price for the first one, so the cost side of your ROI is a real number. You can also book a 30-minute call and talk the numbers through with a founder.
