You probably know the standard calculation: the process costs five hours a week, an hour costs €35, that makes roughly €9,000 a year, the automation costs €12,000, payback after 16 months. We use this back-of-the-envelope version ourselves, and as a first filter it works. As a basis for a decision it is too optimistic, because it omits items on both sides: the cost side is missing operations, maintenance, and your own time, and the benefit side counts every saved hour as cash, which it often is not. Here is the more complete version.

The benefit side: counting hours is the easy part

Measure the time instead of estimating it. Have the people involved note down, over two typical weeks, how much time the process actually eats, including the interruptions and follow-up questions around it. In our experience, estimates are frequently off in both directions: the daily small stuff gets underestimated, the process that feels most annoying gets overestimated.

Calculate with fully loaded cost, not gross wages. With payroll overheads, workplace costs, and downtime, an hour of work in Austria typically costs the company 1.5 to 2 times the gross hourly wage. For office staff, €35 to €50 fully loaded per hour is a workable figure.

The uncomfortable part: saved hours are not saved money

If an automation frees up five hours a week, your payroll drops by exactly zero euros. The employee is still there. The saving only becomes real when the freed time fills up with something useful: backlogged tasks, sales work, or the planned additional part-time hire that no longer needs to happen. In growing companies, that last one is the most common genuine effect: the team handles more volume without growing.

For the calculation, this means: only count saved hours at full value if you can say concretely what will happen with them. If you cannot, count half. That sounds strict, but it prevents the disappointment of an automation that runs perfectly while “nothing got saved”.

The items that are hard to quantify but real

Error costs belong on the benefit side but are tedious to estimate. A transposed digit in an invoice costs a credit note and an awkward phone call. An overlooked order may cost you the customer. Ask the team what actually went wrong in the past twelve months and price those concrete cases, instead of applying generic error rates.

Two items we deliberately leave in the calculation as text rather than numbers: the dependency on individual people that the process has today, and the question of whether the process would still work at twice the volume. Both can matter more than the hourly savings. But translating them into euros would be false precision, and false precision is exactly what this calculation is supposed to avoid.

The full cost side: development is only the beginning

On top of the quoted development price come three items that rarely appear in ROI calculations. First, operations: hosting, updates, minor adjustments, realistically 10 to 20 percent of the development cost per year. Second, your own time: someone in the company has to explain the process, review interim versions, and do the acceptance testing, often several person-days over a project. Third, the rollout: for a few weeks, the old and new way run in parallel, and that costs time rather than saving it.

Also clarify how changes after the project will be priced, because processes change, and an automation that does not keep up loses value every year. This is why we work with a fixed-price option where changes are priced transparently before they are built. However your provider handles it: it belongs in the calculation.

A worked example you can rebuild

Take a fictional plumbing and installation company with twelve technicians. The office manager transfers timesheets from paper and photos into payroll and project billing: a measured six hours per week, which at €40 fully loaded comes to roughly €12,000 a year. The freed time has a clear use, since quoting regularly falls behind, so we count it in full. Add two concrete billing errors from last year with about €1,500 in combined damage. Benefit side: around €13,500 per year.

Cost side: €14,000 development for digital time tracking with a payroll connection, around €2,000 per year for operations and adjustments, an estimated five person-days of internal effort during the project, roughly €1,600 one-off. In year one, that puts about €17,600 in costs against €13,500 in benefits; from year two onward, €2,000 against €13,500. Payback after about a year and a half, and the project earns money from then on. That is a solid calculation, not a spectacular one, and that is what viable automation projects usually look like.

Item	Amount
Benefit: 6 hrs/week of transfer work at €40 fully loaded	roughly €12,000 / year
Benefit: two concrete billing errors from last year	roughly €1,500 / year
Cost: development (one-off)	€14,000
Cost: internal effort, about five person-days (one-off)	roughly €1,600
Cost: operations and adjustments (ongoing)	roughly €2,000 / year
Balance, year 1	€17,600 in costs against €13,500 in benefits
Balance from year 2	€2,000 in costs against €13,500 in benefits

Fictional example from this article

When the numbers argue against automating

There are clear stop criteria. If payback lies beyond three years, the risk is too high that the process changes before then. If the process depends on a system that is due to be replaced, the same applies. And if the benefit side only holds up because soft items were generously translated into euros, it does not hold up. The fact that everyone is automating right now is not a criterion either: according to Statistik Austria’s ICT survey (2024), the share of Austrian companies with ten or more employees using artificial intelligence nearly doubled within a year, from around 11 to 20 percent. That raises the pressure to follow suit, but it changes nothing about your numbers.

A no to a project is not a wasted result. The measurement of real process costs you produce along the way is worth something even when the answer is: keep going manually and redo the numbers in a year.