How to automate business processes as a small business, without it going off the rails

Why automate business processes at all?

Because repetitive work costs more than you think. A task of thirty minutes per working day is about 130 hours a year. Use your own hourly cost; at 45 euros an hour that is 5,850 euros. Per task.

Mistakes weigh more than time. In my own wellbeing platform Holos33, 985 of 1,136 form submissions turned out to be on the wrong list. Nobody had noticed, because no error message appeared. The system worked, just not properly. That did not cost hours, but it did cost enquiries that ended up in the wrong place and were followed up late or wrongly.

So automation gives you three things:

• Less manual work, so time for work that a person has to do.
• Fewer mistakes, because a computer does not get tired on a Friday afternoon.
• Insight into what happens, because every step is recorded.

According to Statistics Netherlands (CBS), 27 percent of companies with 10 to 50 employees used AI in 2025, against 11 percent in 2023. Of the companies that considered it and did not, 73 percent named lack of experience as the main reason. That lack of experience can be fixed. That is what this page is for.

And no, you are not too small. With three people, an hour of manual work costs just as much as with thirty.

How do you automate a process, step by step?

Measure first, then draw, then build and test. Only then go live. Whoever starts with building often builds the wrong thing.

1. Measure the starting point. How many times a week, how many minutes each time, how many mistakes. Write it down.
2. Draw the normal route. What happens in 80 percent of cases? Build only that first.
3. Write down the exceptions. Missing data, duplicate messages, a system that does not respond for a moment. For each case decide: retry, park or stop, and who gets a notification.
4. Build and test with real examples. Ten real enquiries from last month tell you more than a hundred invented ones. Allow two to four hours on your side per process, mostly for supplying those examples.
5. Let someone from the process try to reject it. The colleague who does the work now knows what can go wrong. Someone afraid of losing their job does not look for mistakes but for proof that it does not work. Involve that colleague from the baseline measurement and let them decide what the computer is not allowed to do.
6. Go live in a limited way. First one client, one day, one form. Watch along. Only then the rest.
7. Put a watchdog in place. A check that looks every day whether everything is still running, and raises the alarm if not. Also test whether the alarm really goes off. An alarm that has never gone off is usually not connected.

The full step-by-step plan with baseline measurement, decision limits and handover is in Business Process Automation: A Control-First Guide.

Which software do you use for automation?

For most small businesses: a platform such as n8n, Make or Zapier that ties your existing systems together. No new accounting software, no new CRM. You keep what you have and let it work together.

The difference between the three lies in where your data sits and what you pay as you grow. Zapier and Make charge per action; with more volume that adds up. You can run n8n on your own server, then you pay for the server and not per action, and your customer data stays in house. That last point is the argument I hear most often from accountants and care providers.

More important than the tool: who will maintain it afterwards? An automation nobody maintains often breaks within a year because a connected system changes something.

Where does automation usually go wrong?

With silent failures. An integration that stops without a message, a filter that keeps real customers out, an alarm that was never connected. The system reports nothing, so everyone thinks it works.

Many companies have built something that went quiet after three months. Usually that was not down to the tool.

These are the five causes I see most often:

1. Starting to build instead of measuring. Then you do not know afterwards whether it helped.
2. Starting too big. Five processes at once, of which not one is finished after six months.
3. No owner. Nobody looks at the notifications, so nobody sees that they stop.
4. Forgetting exceptions. The normal route works, but the one customer with an unusual address disappears.
5. Trusting “no error message”.

How to prevent that, with retries, a queue and a watchdog that can fail, is in building an n8n workflow that does not fail silently. What it looked like in practice, with all the figures, is in the Holos33 case. What an alert e-mail in plain language looks like, you can read in the Mercabilis case.

How do you choose someone to build it?

Ask about three things: whether they measure first, what happens when it breaks, and who owns the automation after delivery.

A builder who starts talking about tools straight away has not looked at your process yet. A builder who has no answer to “what happens if the integration is down for ten minutes?” has not been through it before. And if the workflows sit on the builder’s account, you are stuck.

This is what I do myself, and what you can hold me to. I start with a free scan in which I work through one process with you. Then you get a short report with figures, even if the outcome is that automating does not pay off for you yet. If I build something, it runs in your own environment, every integration gets a watchdog that raises the alarm when it stops, and the prices are fixed in advance. More about integrations (for example connecting Exact Online with n8n), administration and finance and leads and quotes.