How to Measure Automation ROI
Someone on your team says an automation will save forty hours a month. You nod, because forty hours sounds like a lot. Then a quieter question arrives. Forty hours of what, at what cost, measured how, and compared to what the automation itself will cost to keep alive? If you cannot answer, you do not have a business case. You have a mood.
Return on investment for automation is not hard to compute. It is easy to compute badly. The usual mistakes are all in the inputs, such as a guessed baseline, an hourly wage instead of a loaded cost, a build price with no upkeep, and savings that depend on people doing something they will not do. This guide walks through each input in the order you would collect it, shows the payback arithmetic with a clearly labeled hypothetical example, and ends with the pitfalls I see most.
It is written for a buyer who thinks in cash flows. You do not need to be technical. You do need to be willing to measure before you buy.
Measure the baseline before anyone builds anything
The baseline is how the work gets done today, in numbers. Without it, any later claim of savings is an argument rather than a result.
For one process, record these for a representative stretch, ideally four weeks.
- Volume. How many items go through, such as invoices, orders, intake forms or reports.
- Touch time. How many minutes of human attention each item takes, counted at the keyboard and not by guess.
- Elapsed time. How long an item waits from arrival to completion, which is often far longer than touch time.
- Error rate. How many items need correction, and how long a correction takes.
- Who does it. The role or roles, because their cost differs.
Time a handful of items with a stopwatch or a screen recording. People remember effort and forget interruptions, so self-reported times usually run low for tedious work and high for work they dislike. Ten real timings beat an hour of estimating in a meeting.
Also note seasonality. A distributor with a heavy December and a quiet February should not annualize a December measurement. In Puerto Rico, hurricane season and the holiday rush can both distort volume, so sample more than one stretch if you can.
Use loaded labor cost, and be honest about what it means
An hourly wage understates what an hour of staff time costs the business. Loaded labor cost adds the employer's share of payroll taxes, benefits, paid time off and often a share of space, equipment and software. A simple way to estimate it is to take the annual cost of a role, including all of those, and divide by the hours the person actually works on tasks, which is less than 2,080 once vacation, holidays, meetings and breaks are removed. Your accountant can give you a defensible multiplier. For hypothetical arithmetic below I use a loaded rate of 32 dollars per hour.
Now the part that finance people care about and enthusiasts skip. Saved hours are only cash savings if something changes because of them. There are three honest outcomes.
- You avoid a hire or overtime you would otherwise have paid for. This is real cash.
- You redeploy the time to work that earns revenue or prevents costs. This is real value, but you must name the work.
- The time goes into slack. The person is a little less busy. This is not a cash saving, even if it improves morale.
Count only what you can place in the first two groups, and label the second as value rather than savings.
Put a price on errors
Manual work produces mistakes, and mistakes cost more than the minutes to fix them. A mistyped invoice amount might trigger a duplicate payment, a late fee, an awkward call with a supplier or a customer who stops trusting your statements. You do not need exact figures. You need a defensible estimate.
For each error type, multiply the following.
- How often it happens, from your baseline.
- The labor to find and fix it.
- Any direct cost, such as a fee, a credit issued or a write-off.
- A flag for any error that could produce a regulatory or contractual consequence, which you should discuss with your adviser and not try to price yourself.
Be careful in the other direction too. Automation does not remove errors, it changes their type. A person mistypes a number. A workflow might apply a wrong rule to a thousand records in a minute. Your after-automation error estimate should reflect that, and your plan should include checks that catch a systematic mistake early.
Count the full cost, including the part that never ends
The build price is the visible cost. The ones that surprise buyers are recurring.
| Cost line | What it covers |
|---|---|
| Discovery and design | Mapping the process and agreeing the rules |
| Build and testing | Configuration or code, plus testing with real data |
| Licenses and usage | Platform subscriptions, per-task or per-record fees, vendor API charges |
| Hosting and monitoring | Where it runs and who watches it |
| Maintenance | Fixes when a vendor changes an API, a form, or a rule |
| Exception handling | Staff time on the items the automation cannot finish |
| Training and change | Time for people to learn the new way |
| Review and governance | Periodic checks that it still does the right thing |
The maintenance line deserves a hard look. Integrations touch other companies' software, and that software changes. Plan for a few hours a month per significant workflow, plus occasional larger fixes. If a proposal states a build price and says nothing about upkeep, ask for it in writing. Ozcorp InfraOps, for what it is worth, starts at 500 dollars per month for the base service with vendor and license charges billed separately, and I mention it only as an example of how recurring costs should be stated plainly rather than buried.
A worked example with illustrative numbers
Everything below is hypothetical. Suppose a small firm processes 400 supplier invoices a month. A clerk keys each one into the accounting system.
Before automation
- Touch time is 6 minutes per invoice, so 400 times 6 gives 2,400 minutes, or 40 hours.
- At 32 dollars per hour loaded, that is 1,280 dollars per month.
- Three percent of invoices contain an error, which is 12 invoices. Each takes 25 minutes to find and correct, so 300 minutes, or 5 hours, which is 160 dollars.
- Total monthly cost is 1,440 dollars.
After automation
- The workflow captures the invoice, extracts the fields and proposes the entry. A person reviews each in an average of 1.5 minutes, so 600 minutes, or 10 hours, which is 320 dollars.
- Errors fall to one percent, which is 4 invoices at 25 minutes, so about 1.7 hours, or 53 dollars.
- Platform and usage fees are 90 dollars per month.
- Maintenance and monitoring take 3 hours a month at a 60 dollar internal or contract rate, which is 180 dollars.
- Total monthly cost is 320 plus 53 plus 90 plus 180, which is 643 dollars.
The comparison
- Monthly saving is 1,440 minus 643, which is 797 dollars.
- Suppose the one-time build, including design, testing and training, is 6,000 dollars.
- Simple payback is 6,000 divided by 797, which is about 7.5 months.
- Over the first twelve months the net result is 12 times 797 minus 6,000, which is 3,564 dollars.
Now stress it. Suppose volume is only 300 invoices a month. Labor falls proportionally, but the platform fee and maintenance stay close to fixed. Rerun the numbers and payback stretches noticeably. Suppose the review step turns out to take 3 minutes. Run it again. The point of the exercise is to learn which assumption moves the answer most, and then to measure that one first.
Also notice what the arithmetic did not include. It did not count faster payment of early-discount invoices, better visibility or lower stress. Those can be real, but add them as separate lines labeled as value, and keep them out of the payback number you present as hard.
Pick the right measure for the decision
Different questions call for different tools.
- Payback period answers how long until the money comes back. It is easy to explain and ignores everything after the break-even point.
- Net benefit over a fixed horizon, such as 12 or 36 months, shows total gain after all costs.
- Return on investment as a percentage is net benefit divided by cost. It is handy for comparing projects of different sizes.
- Net present value discounts future savings to reflect that a dollar later is worth less than a dollar now. For short projects the difference is small, so do not let it slow you down.
If cash is tight, lean on payback. If you are ranking several projects, lean on net benefit over the same horizon for all of them.
Pitfalls I see most often
- Measuring after the fact only. If no baseline exists, savings are a story. Capture it first.
- Counting hours that were never going to be spent. If nobody would have done the work, the saving is zero.
- Ignoring the exceptions. Every automation leaves a residue of items that need a person. Estimate that percentage honestly, and cost it.
- Forgetting adoption. A workflow people route around saves nothing. Budget training and ask the users early.
- Treating the first month as typical. Expect a ramp. Plan for the first month or two at partial benefit.
- Savings that depend on a change nobody owns. If the saving requires staff to stop checking something, name who decides that and when.
- Skipping the review. Revisit the numbers after a quarter and compare them to your estimate. That comparison teaches you more than the original forecast did.
Build a small scorecard you can reuse
For every automation you consider, fill in the same short sheet.
- Process name and owner.
- Baseline volume, touch time, error rate, measured on what dates.
- Loaded rate used and where it came from.
- Expected after-automation touch time and error rate, with the reason you believe it.
- One-time cost and monthly recurring cost.
- Which of the three outcomes (avoided hire, redeployed time, slack) you expect for the saved hours.
- Payback, twelve-month net benefit and the assumption you are least sure about.
- Review date.
Over a year this builds a record of what your forecasts got right, which makes every later decision cheaper.
What to do Monday morning
Pick the one process people complain about most. Time ten real items, count last month's volume and list the errors. Ask your accountant for a loaded hourly rate. Then fill in the scorecard and see whether the payback surprises you in either direction.
If you want a second reader for the numbers, or help measuring a baseline, I would be glad to sit down with you. Sometimes the answer is to automate, sometimes it is to simplify the process first, and sometimes it is to leave it alone. All three are good outcomes when you reach them with evidence.
Discuss your infrastructure requirements
Review the current service offer or schedule a conversation about the scope you need.
Stay in touch
Follow Ozcorp
Leave your details for occasional notes about new articles and Ozcorp updates.