What does AI really mean in your ERP?

“We are going to add AI to our ERP.” It sounds good in the boardroom. But whether you lead an organization of 200 people or are responsible for daily operations, you want to know what concretely changes on Monday morning — when the first email comes in and the first work order needs to be processed. The honest answer: a chatbot on top of your ERP changes very little. Your people already knew how to search. It only gets interesting when AI can not only talk about your ERP, but can also act in it.

From talking to doing

From chat to execute, yes that is possible by using tool calling. You give an AI agent access to the integrations you've had for years — with Exact, AFAS, Business Central, your CRM, your document storage — and let the agent use them autonomously to complete a task. No human intermediary copying a request, but an agent that chooses for itself which action is needed. An example. An email comes in from a customer with a packing slip attached and a request to adjust a quote. Without an agent: someone reads, searches for the customer, opens the project, reads the packing slip, adjusts lines, sends a confirmation. Fifteen minutes, at best. With an agent: the email is read, the packing slip extracted, the customer matched, the change prepared as a draft, a draft reply email ready for review. Two minutes. That final step matters. The agent does the preparatory work, a human approves it. That is where it works in SMEs and industrial companies: not full automation where no one has control anymore, but smart preparation where the human decides.

Why this is only possible now

We have been building ERP integrations for years. That is not new. What is new: a language model can now decide for itself which integration to call to answer a question or perform a task. Previously, every path had to be pre-programmed — if this, then that. Now you can give an agent the command: “resolve this customer question”, and it chooses for itself whether to retrieve project data, analyze a document, or create a new contact. This changes the strategic questions.

For a director: not “which new system should we purchase”, but “how much extra yield is hidden in the systems I already have, if an agent can use them autonomously?”

For the IT or operations manager: not “which integration do we need to build” — you already have those, after all — but “which processes do we let an agent handle, and where do we keep the human in the loop?”

Where you can start tomorrow

Choose one process that takes your people a lot of time and requires a lot of switching between email, ERP, and a third system.

Purchase confirmations, service notifications, requests for quotes.

Choose one, not three. You don't build a good agent for an entire department at once. You build it for one repetitive process and scale up once it proves itself. That is also the honest story behind AI in your ERP. No revolution, no self-managing back office.

But rather: years of integration work that can finally be deployed autonomously, process by process. And that is where the profit lies — measurable, manageable, and for the people who work with it daily, finally noticeable.