Versele Laga AI order intake
From unstructured orders to ERP-ready data, without forcing customers to change
Versele-Laga is a leading international pet food manufacturer, serving a wide and diverse customer base. While many companies push customers toward EDI, the reality is that a large number of customers still place orders via PDFs, Excel files, and emails as forcing change can mean losing business.
Instead of pushing customers to adapt, Versele-Laga chose to adapt their technology, together with Growing Pai.
The challenge & solution
Incoming orders arrived in many different formats and languages and were processed manually by experienced operators. This involved interpreting customer-specific layouts, matching customers and products, and re-entering everything into the ERP system.
With more than 3,000 order documents per month, this manual process was time-consuming, error-prone, and hard to scale.
We built an AI-driven order processing pipeline that automatically ingests PDFs, Excel files, and emails, extracts and validates the order information, and inserts it directly into the ERP system. Human intervention is only required when the system signals uncertainty or to bond with the customer.
Key challenges we solved
Customers are identified based on address information, but in practice this data is often inconsistent, incomplete, or formatted differently across documents and languages. Matching incoming orders with master data was therefore far from trivial. We implemented intelligent matching logic combining document understanding, fuzzy matching, and AI-assisted reasoning.
Each customer orders in their own way, using different layouts, languages, product descriptions, and units such as kilograms, bags, or pieces. The system normalises these variations into standardised ERP order lines, ensuring consistent and reliable product and quantity recognition regardless of how the order was created.
Technology approach
Behind the scenes, the solution uses an internal AI voting mechanism where multiple large language models interpret the same document independently. Their outputs are compared to determine confidence and highlight potential ambiguities for the operator.
The platform is built using Azure AI Foundry models for language understanding and Docling for document parsing, combined with agent validation and feedback loops to continuously improve accuracy.
Still doing manual copy-paste work?
If your teams are manually processing orders, invoices, or other business documents because they are “too complex” or “too unstructured” to automate, we can help.
Growing pAI has proven expertise in building AI solutions that work with real-world data, real constraints, and real business impact.
Let’s explore what this could mean for your organisation: via mail axel@growingpai.com or phone +32 475 54 2216.
