How to Calculate the ROI of Business Automation

Automation ROI is not complicated, but it is easy to inflate. This article gives you the formula we use, shows how to measure the baseline it depends on, works through an example, and lists the mistakes that turn a sensible estimate into a number nobody believes.

The formula

Four lines. Each term is something you can measure or defend.

Monthly gross time value = hours saved per month × loaded hourly cost
Monthly net benefit      = gross time value + error/rework savings + incremental gross profit − recurring software and care costs
Payback period (months)  = implementation fee ÷ monthly net benefit
First-year ROI           = (12 × monthly net benefit − implementation fee) ÷ implementation fee

Loaded hourly cost is salary plus employer contributions, benefits and a share of overhead, divided by productive hours. It is usually 1.3 to 1.6 times the bare hourly wage. Use the loaded figure; the bare wage understates what an hour costs you.

Error and rework savings are the cost of mistakes the automation prevents: re-issued invoices, duplicate records cleaned up, missed enquiries chased, refunds caused by wrong information. Estimate this from your own incident history, not from a vendor’s slide.

Incremental gross profit is the only term that can be zero and often should be. Include it only where you can show a mechanism: faster lead response converting enquiries that were previously lost, for example. If you cannot name the mechanism, leave it at zero.

Recurring costs are platform fees, model usage, hosting and any care plan. They are paid every month, so they reduce the benefit every month.

How to measure a baseline

The formula is only as honest as the “hours saved” figure, and that figure depends on knowing how long the process takes today. Do not guess. For two to four weeks before the build:

1. Count the volume. How many invoices, enquiries, tickets or reports per week? Pull this from the system, not from memory.
2. Time the task. Have the people who do it log start and end times for a sample of at least 20 cases, including the interruptions and the chasing.
3. Count the exceptions. How many cases could not be handled the standard way? These will still need a person after automation.
4. Record the errors. Duplicates, corrections, complaints and re-sends over the period, with a rough cost for each.
5. Note the response time. From trigger to completion. Often the biggest change is not hours but latency.

Write these five numbers down and date them. After launch, measure the same five at 30, 60 and 90 days. That comparison is your ROI evidence; everything before it is a forecast.

Worked example

This is an illustrative example with round numbers, not a client result.

Input	Value	Where it comes from
Hours saved per month	80	Baseline timing minus expected exception handling
Loaded hourly cost	€30	Payroll, loaded
Rework savings per month	€600	Incident log
Incremental gross profit	€0	No defensible mechanism, so excluded
Recurring costs per month	€300	Platform, model usage, care plan
Implementation fee	€6,900	Team package, excl. VAT

• Gross time value: 80 × €30 = €2,400/month
• Net monthly benefit: €2,400 + €600 + €0 − €300 = €2,700/month
• Payback: €6,900 ÷ €2,700 ≈ 2.6 months
• First-year ROI: (12 × €2,700 − €6,900) ÷ €6,900 ≈ 370%

Now halve the hours saved to 40 and the picture changes: net benefit €1,500, payback 4.6 months, first-year ROI about 160%. Still worth doing, but a different decision. This sensitivity is why the ROI calculator lets you change every assumption and shows the formula.

Common mistakes

Counting saved hours as cash. Eighty saved hours do not appear on the bank statement unless you reduce overtime, avoid a hire or redeploy the time to work that earns money. Saved hours are capacity. State what you will do with the capacity, or discount the figure.

Ignoring adoption. A workflow the team routes around saves nothing. If half the cases still go the old way, halve the benefit. Training, a gradual rollout and a visible approval queue are what move adoption, and they belong in the plan.

Ignoring exceptions. Automation handles the standard cases. The rest still need a person, and they are usually the slow ones. If 10% of cases are exceptions that take three times as long, the hours saved are less than volume × time suggests.

Forgetting recurring costs. Platform fees, model usage and maintenance are small individually and permanent collectively. Leaving them out overstates net benefit every month.

Using vendor benchmarks as your baseline. “Companies save 30%” tells you nothing about your process. Use your own numbers or say you do not know yet.

Measuring only once. A forecast is not a result. Without a 30/60/90-day comparison, nobody can tell whether the project paid off.

A guardrail for value-based pricing

Some providers price on value rather than effort. That is legitimate when the value is measurable, and a trap when it is not. The guardrail we apply: a project price should sit at roughly 15–30% of defensible first-year value, provided it still covers delivery cost, risk, third-party costs and support. “Defensible” means the value is computed from a measured baseline with the formula above, with incremental profit excluded unless there is a mechanism. If the baseline cannot be measured, value pricing should not be used at all; a fixed-scope package is fairer for both sides.

Applied to the example: first-year net benefit of €32,400 would put a value-based price between about €4,900 and €9,700. A €6,900 fixed fee sits inside that range, which is one way to sanity-check a quote.

Risks in the estimate

Every figure above is an estimate until the 90-day comparison. The hours figure is the most fragile, followed by adoption. Treat the payback period as a planning range, not a promise, and be suspicious of any provider who presents it otherwise.

Next actions

• Pick one process and record the five baseline numbers for two weeks.
• Put them into the ROI calculator and test the sensitivity by halving the hours.
• If the payback is under a year at the conservative figure, read what AI automation costs and compare packages on the pricing page.