← All cases

AI Empowers · Pet food manufacturing

AI order intake at Versele-Laga

Orders arrive as PDF, Excel and email in every layout and language. AI turns them into ERP order lines.

  • Manufacturing
  • Order automation
  • Document intelligence
Client
SectorPet food manufacturing
CategoryAI Empowers
StatusIn production since September 2025
80%of order emails handled automatically

The problem

Orders arrived by email as PDFs, spreadsheets and scans in many layouts, languages and document types, and were typed into the ERP by hand.

The solution

An AI pipeline extracts each order, matches the customer (fuzzy matching plus AI), normalises units and writes ERP order lines. People only review when the system is unsure. Adapt the technology, not the customer: no one had to switch to EDI.

The outcome

About 80% of order emails are handled automatically and more than 5,000 orders have been processed, with human review only on the uncertain ones. Three teams use it, and it runs in four countries.

5,000+orders processed so far
4countries live: Belgium, the Netherlands, France and Germany
>80%accuracy on live orders

The full story

  1. Where it started

    About two thirds of all orders, some 67,000 order lines a month, arrived as e-mails, PDFs, Excel and Word files, with a different layout per customer and often per sales rep. Staff keyed them line by line into the in-house order system.

  2. Proof first

    We kicked off in January 2025 with a three-week proof of concept on real orders. After fine-tuning, 16 of 20 test orders came out fully correct, and the system itself flagged 98% of the remaining errors.

  3. How it works

    A classifier picks the orders out of the inbox. Several PDF parsers read each document in parallel, then three LLM agents extract the order independently and vote. Products are matched on EAN and article number, even when a customer quotes the wrong EAN, and customers on their delivery address. Anything uncertain is flagged on the order for a person to check.

  4. In production

    Connected to the order system since September 2025, it has been live for over a year. Three teams in Aalter, Deinze and Maastricht use it, it handles orders in several languages and it avoids typing errors. OCR for scanned PDFs followed in October 2026, and the rollout has since reached Hungary. Users validate the uncertain fields and correct exceptions, and that feedback keeps improving the system.

Tech

Azure AI FoundryAzure Document IntelligenceAzure AI SearchDoclingMulti-LLM votingAgent validation loops
Facing something similar?We would shape your version around your own data.

Ask Pai

Ask Pai about this case

Pai is an AI. It only knows what is written on this page.

Baby Pai · AIHi, I'm Baby Pai, an AI assistant. Ask me about the Versele-Laga case.

Pai is an AI. Answers come from this case page. Don't share sensitive data.

Chats are kept for 30 days unless you switch that off in Privacy and choices.