Subsidies for your AI project?

Check your eligibility
Customer story
9 October 2026

Sibelco classifies incoming orders with over 95% accuracy using an AI agent

Sibelco logo

In our words •

Sibelco, the Antwerp-based global materials group, receives customer orders in a central mailbox. The same mailbox also receives transport orders, complaints, invoices and other documents. Customers often use their own product names, codes and numbering. Before an order could be processed, Customer Service first had to work out which documents were actual purchase orders.

Together with Sibelco, CROPLAND built an Order Intake Agent that reads and classifies every incoming document, including technical specification sheets. The agent runs on an OpenAI language model and works with the systems Sibelco already uses. Incoming emails are logged as cases in Salesforce and each attachment is analysed separately. A second model filters out irrelevant files, and a check prevents the same document from being processed twice. When the agent is not confident enough, it hands the document to an employee.

The results:

  • Classification accuracy above 95%, up from 80 to 85%
  • About half an hour of manual work saved per order
  • More time for order processing, follow-up and quality control

The project started with an AI discovery session. Over the years, many business rules had grown and partly lived in the heads of the Customer Service team. Mapping processes, decision trees and exceptions together with the users made those rules explicit. A next phase looks at letting the agent act directly in the ERP system.

How much knowledge in your organisation still lives only in people’s heads?

Context

The case was picked up by several Belgian business and technology media in October 2026.

Published in

4 outlets

More in the press

The founders of the Serso Gravelbike Team with Geert Vromman at the CROPLAND office
Partnership
21 September 2026

Serso Gravelbike Team uses AI behind the scenes, with support from CROPLAND

CROPLAND sponsors the Serso Gravelbike Team and shares AI know-how, so a small team spends less time on communication and administration.
Appeared in 1 outlet
Opinion
17 September 2026

What if we no longer understand our own work?

AI agents take over decisions, not just tasks. Geert Vromman explains why understanding your own processes becomes more important, not less.
Appeared in 1 outlet