Modern API with Artificial Intelligence for Reading and Posting Cost Invoices

A company managing complex projects across multiple markets, simultaneously handling dozens of investments in various countries. Each month, they processed thousands of cost invoices in different languages and formats. 

The client received thousands of invoices each month from vendors around the world—in a wide variety of languages, formats, and currencies. Differences in the structure of the documents (e.g., each vendor displayed line items and totals differently) made automated data extraction unreliable. OCR could extract text, but retrieving specific data (such as net/gross amounts, taxes, or project numbers) often required manual intervention by accountants, especially when the invoice was in Greek or Hungarian.

Accurately assigning costs to the correct internal projects was also critical—vendors did not always clearly indicate project or contract numbers. The client needed a solution that could fully automate invoice processing—regardless of format or source—and integrate with their existing Document Management System (DMS).

The client opted for a proprietary Artificial Intelligence solution combining OCR and NLP, tailored to their needs.

• We developed a web-based API that accepts invoice files (scanned PDFs or even photos/faxes). Each invoice is first converted into text using OCR technology and then analyzed by a specially trained AI model.
• The model was trained to identify key information regardless of format—it can locate sections with seller and buyer details, extract line items and values, detect VAT rates, and find project numbers (by recognizing character patterns based on predefined templates, whether in item descriptions, headers, or notes).
• Users (accounting team) can define what data should be extracted via an additional interface. This is especially important as business requirements evolve.
• After processing, the API returns structured data in JSON format—ready to be imported into the internal financial system. In this case, the results were loaded into the company’s DMS, which managed approval and accounting workflows.

• Automated up to 90% of the work related to data extraction from invoices—manual verification is now only needed when there’s uncertainty in the data readout.
• Reduced invoice posting time—from several days to just minutes after the document is received.
• Significant cost savings—the cost of processing a single invoice dropped several times compared to existing SaaS tools.
• Full compliance with client requirements—a flexible API adapted to varying invoice formats and international contexts.
• Improved financial reporting—the finance team had access to most cost data before month-end.
• Multilingual and multi-format invoice handling—the AI model successfully processed even less common languages (e.g., Greek, Hungarian).

Our custom-designed solution, fully integrated with the DocuWare system, turned out to be eight times more cost-effective than a comparable offer from a global e-accounting service provider. The client gained full functionality at a fraction of the cost, without compromising on security or process automation.

By implementing its own AI API, the company eliminated the need for flawed, expensive third-party tools. Automation made the process faster, cheaper, and more accurate, giving the finance department real support in managing complex international project accounting.