Which business processes can you automate — and which should you leave alone?

A business process is a good candidate for automation when it happens often, follows the same steps, runs on rules you can write down and is already digital: sorting customer questions, reading invoices, moving data between systems. Judgement calls, exceptions and personal contact are better left with people. Goudbeek in Almere therefore starts with the process, not the tool.

Below: how to recognise such a process, the examples that turn up in many businesses, what to keep with people, when AI adds something and how to decide where to start.

What makes a process worth automating?

Four traits predict fairly well whether automation makes sense. The more of them hold, the fewer surprises.

It happens often. A task that comes back every day or every week is worth handing over. Something that happens a few times a year is usually better served by a good checklist.

It runs the same way every time. Same steps in the same order, with the same result. If everyone does it their own way, agree on one way first. Only then is there anything to automate.

It follows rules. You can write down when something goes one way or the other: above this amount to the director, this supplier always to that cost centre. The better you can spell out the rules, the better a system can follow them.

It is already digital. The information lives in a mailbox, a form, a webshop or accounting software. Paper can work too, but it adds a step: reading the document reliably first. That is what document processing with AI is for.

Which processes do businesses automate in practice?

Good candidates are often not in the core work but in the work around it. A few examples.

Customer questions. Picture a webshop that gets the same questions every day about delivery, returns and order status. A system can recognise those questions, answer them from the shop's own help pages and order data, and pass anything doubtful to a colleague with a draft reply. An AI agent for customer service takes over the queue, not the conversation with the difficult customer.

Invoices and documents. Picture an accountancy firm retyping its clients' purchase invoices. Reading the supplier, date, amounts and line items is textbook automation work: the system checks the totals, prepares what it is sure of and puts anything uncertain on a review list.

Data between systems. Picture a heating and plumbing installer where every new job means the customer into the CRM, the appointment into the schedule, a work order, and later an invoice in the bookkeeping. The same details, typed in again and again. That copying is exactly what automating the work between your systems removes.

Following up leads. An enquiry from the website lands in a shared inbox and sits there because nobody feels it is theirs. Logging it, assigning it to the right person and sending a reminder if nothing happens lends itself well to automation. Every enquiry then has an owner.

Reporting. If you paste figures from three systems into a spreadsheet every Monday, you are doing work a dashboard for your processes can do continuously. You no longer see the numbers once a week, but whenever you want.

Which processes are better left alone?

Not everything that can be automated should be. Three kinds of work generally stay with people, at most with help from a system.

Judgement. A quote for an awkward renovation, a goodwill gesture for a loyal customer, deciding whether a complaint is fair: someone is weighing things that do not fit into rules. A system can gather the information and suggest a draft. The decision stays with a person.

Exceptions. Every process has cases that are slightly different, and automating all of them makes the system complicated and fragile. Automate the ordinary cases instead, and have the system stop and hand over the moment something does not fit. A system that guesses is worse than a system that stops.

Relationships. The conversation with an unhappy customer, the call to a long-standing supplier, the meeting that saves a partnership: there, the human contact is the point. Automate the work around it so there is more time for it.

Decisions about people come with rules of their own. The GDPR (Article 22), for example, limits decisions based solely on automated processing that produce legal effects for someone or similarly significantly affect them. What that means for your situation is in the official text, linked below this piece.

Do you need AI to automate a process?

Often not. A lot of automation is simply an integration: when an order arrives in the webshop, create the customer in the bookkeeping. No language model involved, just two systems talking to each other.

Sometimes a good form is enough. Say customers report faults in loose emails and someone has to chase every missing detail. A form with the right required fields solves much of that before anything is automated at all.

AI earns its place where the input is unstructured: an email in someone's own words, a scan with a changing layout, a question that can be asked in any number of ways. There a language model can read and sort. The rules around it, such as what it may do and when it stops, are still ordinary software.

A useful order of questions: can the process be simpler? Would a form or template help? Would an integration solve it? And only then: does AI add anything?

How do you choose which process to automate first?

Start where the work actually gets stuck, not with the tool you would like to try. Ask your team which job eats the most hours and where things go wrong most often.

Map that process as it really runs, including the exceptions everyone knows and nobody wrote down. That kind of process analysis of how your team works is useful on its own: sometimes a step turns out to be redundant or duplicated.

Then pick one process that happens often, has clear rules and where a mistake is limited and reversible. Do not automate a process that does not work; you only make the mess run faster.

Build one working version on real data and run it alongside the manual way before anyone stops doing it by hand.

What happens when an automated step fails?

Every automation gets stuck somewhere sooner or later: a system is briefly unreachable, a supplier changes its invoice layout, a customer types something odd. What matters is what happens next.

Decide per step, in advance, whether the system retries, stops or asks a person. A temporary outage can be retried. An invoice that does not add up goes to a review list. Anything irreversible, like a payment or a deletion, waits for someone by default.

Make sure you can see what happened: what came in, what went through, what stopped and why. And make sure the whole thing can be switched off without the rest of the business grinding to a halt. Then a failure is a message in the morning, not a surprise at the end of the month.

How does Goudbeek approach process automation?

We start with the process, not the technology. Together with your team we map how the work runs now and where time or quality is lost. Then we pick the simplest thing that works: that might be an AI agent, but it might just as well be an integration, a form or a dashboard.

After a short intake we agree a fixed price. We build one working prototype on real data before going further, and write down in advance what the system may see and do, and what always waits for a person. After that it goes live on our managed hosting, from €39 a month, and we adjust it based on real use.

Wondering which of your processes qualify? Read how we analyse and improve business processes, or tell us which job takes the most time. We reply within one working day.