How to Calculate Automation ROI: Costs, Savings and Payback Period

Calculating automation ROI looks easy: hours saved times hourly cost, divided by the project price. Almost every online calculator uses that formula. That is also why almost every one of them promises returns that never materialise.

The formula is not the problem. What it leaves out is. Below we work through a real-world example twice, first the naive way and then the honest way, and show where the gap comes from.

The ROI and payback formulas

Two metrics matter to a decision-maker, and they answer different questions.

ROI (return on investment) tells you how much you gain for every euro spent over a set period:

ROI = (cumulative benefits − investment) ÷ investment

The payback period tells you how many months it takes to recover what you spent:

Payback = upfront investment ÷ net monthly saving

For an SME, payback usually matters more than ROI. A 300% five-year ROI is of little use if cash flow cannot wait three years to get the money back.

For horizons up to two years, simple payback is enough. Beyond that, discount future cash flows: a euro saved four years from now is worth less than a euro today.

What goes into the investment

The investment is more than the implementation quote. It covers everything you pay once:

• Implementation: analysis, build, integrations, testing and go-live.
• Your team's time: whoever knows the process has to explain it, supply real examples and test. Those hours come out of operations.
• Data clean-up: duplicate customers or inconsistent product codes will trip up any automation.

We covered the order of magnitude of each item in our article on how much process automation costs for an SME. Here we look at the next step: what happens to that cost over time.

What counts as a saving, and what does not

On the benefit side, count only what you can measure:

• Hours freed up, measured before and after, not estimated from memory.
• Errors avoided, but only if you know how many happen today and what each one costs.
• External costs avoided, such as overtime or temporary staff.

Everything else ("more agility", "better customer experience") may be real, but it stays out of the calculation. If a project only stacks up on benefits you cannot measure, it is not yet a good project.

On the cost side, the gross saving has to cover the platform, AI usage and maintenance every single month. This is where calculators usually go wrong: they treat automation as a one-off cost when it is a running expense.

Worked example: 480 sales orders a month

A distributor receives 480 orders a month by email. Each takes 5 minutes to read, check stock and enter into the ERP.

Current cost: 480 × 5 minutes = 40 hours a month.

In 2025, according to Eurostat's hourly labour cost statistics, an hour of work cost Portuguese employers €19.40 on average, including social charges. The EU average was €34.90. Our article on how much manual processes actually cost sets out the method for your own processes.

40 hours × €19.40 = €776 a month.

The naive calculation

Assume an implementation quote of €4,000. This is a hypothesis for the example, not a price list.

The naive version assumes all 40 hours disappear and that the automation costs nothing once delivered:

• Payback: €4,000 ÷ €776 = 5.2 months
• Three-year ROI: (36 × €776 − €4,000) ÷ €4,000 = 598%

These are the figures that tend to appear in a sales deck.

The honest calculation

Now with what actually happens.

Not everything gets automated. Some orders arrive with wrong references, ambiguous quantities or special requests. We assume 15% go to manual review: 72 orders × 5 minutes = 6 hours.

The rest still needs oversight. Even clean orders get a quick check, around 30 seconds each: 408 × 0.5 minutes = 3.4 hours.

Hours that remain: 9.4. Hours actually freed up: 30.6.

Gross saving: 30.6 × €19.40 = €594 a month.

Running costs:

• Platform: the Starter plan on n8n's pricing page costs €20 a month billed annually and includes 2,500 executions. 480 orders fit easily, since one execution is one full run of the workflow.
• AI: according to Anthropic's pricing, Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens. At an estimated 3,000 input and 500 output tokens per order, each one costs about 1.1 US cents. Around €5 a month.
• Maintenance: we assume €100 a month to handle errors, adjust rules and react when a customer changes their email format.

Total running cost: about €125 a month.

Net saving: €594 − €125 = €469 a month.

Real investment: add 20 hours of team time during the project (20 × €19.40 = €388) to the €4,000. Total: €4,388.

Simple payback: €4,388 ÷ €469 = 9.4 months.

The adoption curve delays the return

There is one more effect no calculator includes. In the first few weeks, the team does not trust the output and checks everything.

If only half the gross saving is captured in the first three months, the net saving over that period drops to about €172 a month.

With that slow start, the investment is recovered around month 12. Three-year ROI comes out at 264%.

Still a sound investment. But payback is more than twice what the naive version promised, and ROI is less than half. Whoever signed off on five months will think something has gone wrong by month six, when everything went exactly as expected.

Hours freed up are not euros saved

This is the most overlooked point in any automation ROI calculation.

Thirty hours a month is roughly 20% of one full-time role. Nobody leaves because of it, and the salary is still paid.

Hours only turn into money in one of four situations:

1. You avoid a hire that growth would otherwise require.
2. You cut overtime or temporary staff you pay for today.
3. You free time for revenue-generating work, such as account management or proposals.
4. You reduce errors with a direct cost, such as credit notes or returns.

If none applies, what you gained is capacity, not savings. It may still be worth it, but it is a different decision and should be presented as one.

Sensitivity analysis: where ROI is decided

A single payback figure gives false certainty. It is more useful to know which variable could break it.

If the exception rate is 30% instead of 15%, net saving falls to €364 and simple payback rises to 12 months. Still worthwhile.

If volume is 120 orders instead of 480, gross saving is €148. Running costs barely fall, because the platform and maintenance are fixed. That leaves €27 a month, and the investment would take over 13 years to recover.

The lesson: the exception rate matters, but volume matters far more. Below a certain volume, automation does not pay for itself, however good the build.

Measure before you calculate

All of this depends on two numbers most companies do not have: how long each run takes today and how often it fails.

Without them, ROI is an opinion with decimal places. That is why we recommend measuring for at least two weeks before touching the process, as explained in our guide to building an automation pilot that proves value.

Then redo the calculation after three months in production, with real exception and oversight figures. That is the ROI that counts.

The decision

Before approving your next project, ask for the automation ROI worked out this way, and put three questions to whoever proposes it. Does the saving account for exceptions and oversight? Are running costs included? And what will the freed-up hours be used for?

If nobody can answer the third one, what work in your company should those hours be doing?