It hides in small pieces
Ten minutes retyping an order, fifteen checking an invoice, a quick call to see whether a request came in. Nobody sees it as a project, so nobody adds it up. That is exactly why it stays.
Retyping, checking, chasing, fixing mistakes. Work out in one minute what it costs each year, using your own figures.
Built for clients including Tata Steel, M&G Investments, Digistore24, Toppers in Concert, Bar-le-Duc and Loogman
The same method we use in the AI Opportunity Audit: hours multiplied by what those hours cost, plus time spent fixing mistakes.
Manual work costs you the number of hours per week, times the number of working weeks, times what an hour of work costs you. Three people who each spend 6 hours a week retyping orders, plus 1 hour a week fixing mistakes, cost you €39,330 a year at 46 working weeks and €45 an hour. That is 874 hours of work nobody enjoys.
The hourly rate is where most sums already go wrong. People calculate with gross pay and forget the employer costs on top. According to CBS (Statistics Netherlands), an employer in the Netherlands paid on average €45.00 per hour worked in 2024, including holiday pay, social security contributions and other employer costs (CBS, 15 October 2025). If you do not know your own figure, enter that average under "other", or pick the example amount closest to it. If your team is mostly outside the Netherlands, your local figure will differ.
The calculator works with 46 working weeks a year: 52 weeks minus holidays and public holidays. Otherwise it only uses your own numbers. There is no assumption about savings in it, and that is on purpose.
People almost always forget these in a quick estimate:
What I would not do: calculate with gross monthly pay only. Then manual work looks cheaper than it is, and every solution looks expensive.
Below is a worked example with three tasks in a fictional sales support team. You fill in the calculator once per task. Together the three tasks cost 1,219 hours and about €54,900 a year. The most expensive task is not the one the team complains about.
| Task | People | Hours per week pp | Fixing mistakes per week | Hours per year | Cost per year |
|---|---|---|---|---|---|
| Retyping orders | 3 | 6 | 0 | 828 | €37,260 |
| Entering purchase invoices | 1 | 5 | 0.5 | 253 | €11,385 |
| Producing the weekly report | 1 | 3 | 0 | 138 | €6,210 |
| Total | 1,219 | €54,855 |
Example, not a client case. Method: 46 working weeks and €45 an hour, as the live calculator works.
What costs the most here? Not what you think. In this example the team mainly grumbles about the weekly report, because it has to be done by Monday morning. Yet retyping orders costs six times as much. The invoices show why fixing mistakes counts: half an hour a week turns 230 hours into 253. And the report costs €6,210 a year. For a task like that, I would first check whether your reporting tool can already do it, before having anything built.
Keep a tally for a week. Everyone who does the task makes a mark each time and notes how many minutes it took. After five working days you have real numbers instead of a feeling. After all, people rarely estimate the time for boring work well: sometimes too high out of irritation, sometimes too low because it is "only a moment".
Counting in minutes is often easier. If someone handles 40 invoices a week at 4 minutes each, that is 160 minutes, so almost 2.7 hours a week. That is the number you enter as hours per person.
A tally week of 5 working days is enough for tasks that come back every week. For monthly work, such as the month-end close, you tally for a whole month instead. Also pick a normal week. The week before Christmas or the middle of a reorganisation gives strange numbers. An example tally sheet is in automating repetitive tasks.
Manual work goes unnoticed because it is spread over many people and many small moments. Added up over a year, it is often the biggest cost nobody has in view.
Ten minutes retyping an order, fifteen checking an invoice, a quick call to see whether a request came in. Nobody sees it as a project, so nobody adds it up. That is exactly why it stays.
Anyone moving data by hand from one system to another makes mistakes. Those mistakes cost time again: working out what happened, correcting it, explaining to the customer what went wrong. That rework belongs in the sum, which is why the calculator asks for it separately.
When the work comes up often, the rules are largely fixed and the current work costs more than building and running a solution. If the rules are fixed and the input is clean, an integration is often enough. Only when language or judgement is involved does an AI model come in.
The calculator gives you an order of magnitude. In the AI Opportunity Audit we work it out precisely for each process, with the people who do the work, and set the cost of building and running a solution next to it. If the sum comes out negative, the report says so.
In the audit we look at the whole process, from the email that comes in to the entry in your accounts. Often the most expensive step then turns out not to be the step everyone complains about. Sometimes the manual work even sits in a spreadsheet model: see a quoting tool for which we turned an Excel calculation model into code. The step-by-step plan from measuring to go-live is in automating business processes with AI.
Against what the work costs now, you set the costs of a solution: the build, the maintenance and the licences. On top of that comes the time of your own people to explain the process and check the outcome. And no solution takes over everything. Exceptions remain, such as an invoice without an order number, and so do checks you deliberately leave with a person. Anyone who promises you 100 percent has not seen your exceptions yet.
| Cost item | One-off or ongoing | What it consists of |
|---|---|---|
| Build | one-off | working out the process, connecting your systems, testing with real examples |
| Maintenance | ongoing | updates of connected systems, changes when your process changes, someone who picks up outages |
| Licences and usage | ongoing, per month or per use | the AI model, a workflow tool |
| Your own time | mostly at the start | explaining the process, checking outcomes, handling exceptions |
Maintenance is forgotten most often. An AI solution does not keep running by itself: systems get updates and rules change. Besides, a saved hour only becomes a saved euro if you know what that colleague will do with the hour. If you put it towards customers, sales or something that is currently left undone, it counts. Otherwise the gain leaks away into small jobs.
A business case for AI only adds up if what you win back each year is higher than build, maintenance and licences together, over the period you use the solution. What such a solution costs, you can read under what an AI solution costs. How the split between routine and exception works is explained in what is an AI employee. For a sum over several years there is the AI ROI model (in Dutch).
If the calculator comes out below 100 hours a year, first check whether your current software can already do it. A small task that comes back once a month almost never earns itself back, however clever the solution. That may sound odd from a company that builds AI. Still, it is part of the job.
My advice for tasks that only come back a few times a year, such as an annual stocktake: leave them as they are, or capture the steps in a good checklist. That serves you better than software that sits idle all year and still needs maintenance.

Send us your result and we'll go through it with you: which part automation, AI or an integration can take over, and which part it can't.
The number of people multiplied by the hours they spend on the work each week, plus the hours spent fixing mistakes. We multiply that by the hourly cost including employer costs and by 46 working weeks a year. It only uses your own figures; there is no assumption about savings.
Because how much you win back differs per process. Sometimes an integration between two systems takes over almost all of the work; sometimes a lot of exception handling stays with a person. We work that out in the AI Opportunity Audit instead of guessing it with a slider.
Revenue that slips because work isn't picked up, or is picked up too late: a quote that goes out a day later, a customer query left waiting, an invoice sent late. That isn't in the calculator, but it does count in the audit.
What an hour of work costs the organisation: gross salary plus employer costs such as pension, social security contributions and holiday pay. If you don't know exactly, pick one of the example amounts; the result is meant as an order of magnitude.
When the work comes up often, the rules are largely fixed and the current work costs more than building and running a solution. Work where every situation is different, or where a mistake has serious consequences without anyone checking, is better left with a person.
A five-working-day review of where AI, automation or new software genuinely adds value in your organisation. You provide an intake and three to five interviews; the report shows what things cost now, which three opportunities are worth the most and what we would not do.
According to CBS, an employer in the Netherlands paid on average €45.00 per hour worked in 2024, including employer costs. For your team it may be higher or lower. Enter your own hourly cost in the calculator if you know it.
What the manual work costs now, a cautious estimate of what you win back and all the costs of the solution: build, maintenance and licences. Set them side by side over the same period. The calculator gives the first figure; what you win back and what the solution costs, the audit works out per process.