AI Order Capture at ASTORtec: PDF orders transferred automatically to the ERP

ASTORtec supplies companies from industry and construction with adhesives, seals, rolled materials, stamped parts and other technical products. Many B2B orders arrive by email as PDF attachments – frequently generated automatically from the customers’ ordering systems.

These documents differ considerably. Individual article numbers, differing quantity and packaging units, delivery dates and subsequent corrections must be recognised and assigned correctly.

Until now, employees checked each order manually and transferred it to the ERP. This took an average of around ten minutes per order – considerably longer for extensive PDFs. In addition, there was a risk of transfer errors.

The solution was to make differently structured orders usable for automatic processing. The focus was on:

• Reduce entry work: adopt order data from PDFs instead of typing it in.
• Standardise and check data: take customer formats and special order logic into account.
• Connect Europa3000: continue processing checked data directly in the existing ERP.
• Retain control: specifically hold back orders with deviations or missing information for review.

The Evoya Order App combines AI-powered document processing with fixed business rules and an ERP integration. The process comprises four steps.

1. Detect orders and read PDFs: Incoming order emails are forwarded to the application. It first checks whether an actual order is present. Mistral OCR then reads the PDF attachments. Document recognition captures text as well as table and line item data and prepares it for further processing.

2. Interpret and structure order data with AI: Language models interpret the extracted information and bring it into a uniform structure. Customer-specific prompts provide additional guidance on how particular order formats should be read. This allows the application to handle individual article numbers, differing designations and special order logic. The AI therefore does more than plain text recognition: it classifies the content for the subsequent business process.

3. Validate data and hand it over to Europa3000: Defined validation rules apply before the ERP transfer. They check, among other things, quantities and packaging units, delivery dates, article assignments and further customer-specific criteria. If all criteria are met and automatic processing is enabled for the customer in question, the application creates the order directly in Europa3000.

4. Hold back unclear cases for manual review: In case of deviations or missing information, the order remains in the Evoya Order App. Employees can review, correct and approve the data via the web interface. This eliminates full manual entry while control over exceptions is retained. AI interprets the documents; fixed rules and human approvals control further processing.

Orders contain confidential business information and may include personal data. That is why the question of where and how the AI processes data is part of designing the solution.

For ASTORtec, Evoya AI uses Mistral on sovereign EU infrastructure. The processing location of the AI is thus deliberately chosen as part of the architecture. For comparable projects, Evoya AI offers further operating models:

The choice depends on the customer’s requirements. Beyond the location, access rights, retention, involved service providers and contractual arrangements must also be considered. This way, data privacy is not reduced to a hosting label but is assessed in the concrete data flow.

For more information, see Security and privacy at Evoya AI and On-premises AI solutions.

Orders that previously required an average of around ten minutes of manual entry are now either processed automatically or checked and approved in one to two minutes. These figures refer to the manual processing effort, not to the total technical throughput time.

This gives ASTORtec concrete benefits:

• Less data entry: employees no longer have to type in order data in full.
• Targeted handling of exceptions: attention is mainly required for orders with deviations or missing details.
• Fewer transfer errors: manual transfer of article numbers, quantities, prices and delivery dates is reduced.
• Continued use of the existing ERP: Europa3000 remains in use and is extended with AI-powered processing and a web interface.

The approach is not limited to orders. Wherever information from incoming PDFs is read, checked and transferred to a business system, a comparable AI solution can be applied. The basic principle remains the same: receive documents, extract and structure content, apply business rules and pass checked data on to the next process.

Possible further use cases are:

These applications describe the potential of the approach beyond the ASTORtec project. Evoya AI adapts document processing, validation rules and approvals to the respective process.

ERP integration beyond Europa3000

A comparable solution can also be connected to other ERP systems, provided suitable interfaces are available. These include, for example, SAP S/4HANA, Microsoft Dynamics 365 Business Central, Abacus, Sage and Odoo. The integration is tailored to the specific system, its version and the required functions. It is not just about transferring data: the integration must enable the actions required by the process, such as creating an order.

For companies in Switzerland, Germany and Austria, this opens up a practical way to keep using existing systems while reducing manual work in between. A clearly delimited document process can serve as an entry point and later be extended with further use cases. More about this approach: Custom AI applications from Evoya AI.