Routine work runs automatically while you handle exceptions.

An order comes in by email. Someone enters it in the ERP, creates an order confirmation, sends it to the client, and adds a line to the planning. Four times the same thing, every day, for every order. Not difficult, but work that piles up as soon as that one colleague is out sick. We make sure those steps run by themselves and that your team only looks at what deviates.

The steps you repeat every day

Almost every company has a handful of them. A request that has to go from the contact form to the CRM. A timesheet that is retyped into invoicing at the end of the week. A supplier who asks for a confirmation as soon as the goods arrive. On their own, they are small things. Added up, it is a day a week, spread over three people, and nobody sees it because it has always been there.

We start by writing things down. We follow a workday and note which actions go the same way every time: the same data, the same systems, the same order. That is the list we work with. What is not on it, because it needs a person's judgment every time, we leave alone for now.

One workflow to a prototype within ten business days

We do not tackle everything at once. From the list, we choose together the workflow that costs the most time or goes wrong most often, and we build that one first. Within ten business days, a prototype runs that performs those steps itself, on your own systems and with your own data.

You let that prototype run alongside the current way of working for a while. Your colleague still does the work by hand, and the workflow does it next to them. Every day, you compare the results. If they match, the hand comes off the wheel. If they do not, we see exactly at which step and adjust the rule. Then the next workflow is up. The integrations needed for that, between the CRM and the accounting system for example, we build as System integration, so the second workflow can reuse them right away.

Rules where possible, AI where needed

Most steps are simply a rule. If the order amount is below the threshold, approve. If the client is abroad, use the other VAT code. We write such rules out in code you can read, because a rule you cannot check is one you eventually stop trusting.

Sometimes the input is too messy for a rule. A quote request in free text, a delivery note as a photo in an email attachment, a complaint the order number still has to be pulled from. For that, we use a language model, on a server in the EU, and only for that one step. The model extracts the data, the rules do the rest. If the model is not sure, the workflow sets the step aside for a colleague instead of guessing. If you want the automation to handle several tasks in a row on its own and make its own decisions along the way, you are closer to AI agents for the back office. We tell you that in the intake.

What happens when a step fails

A workflow that stops silently is worse than no workflow, because then everyone thinks it is taken care of. That is why every workflow we build keeps a log: which step, when, with which data, and what the result was. If a step fails, because the accounting system is briefly unreachable or a supplier leaves a field empty, the workflow stops at that point, keeps the queue, and sends an alert to the person who needs to know. As soon as it can, it picks up where it left off.

What we build is yours. You get the code, the rules in plain language, and the documentation, and the data stays within the EU. You can take over the maintenance yourself or leave it with us under Software maintenance. Not sure whether a process can be automated at all? Then we look at that in the free intake, and if the answer is no, we tell you right away.