Client

A capital group specializing in real estate management across Poland and Western Europe.

Expectations

The client sought to alleviate the daily manual workload of the primary secretariat staff involved in bulk processing of data from incoming invoices and accounting notes. They desired the implementation of sophisticated tools to streamline this process, and the automatic (sans employee intervention) transfer of crucial accounting data to an Electronic Document Management (EDM) system.

Project

The client decided to implement an effective tool designed to ease the everyday work of several secretariat employees. The project encompassed a stage of thorough analysis of processed documents (including their types, data structure contained in them, the image quality of scanned paper invoices, entry channels to the process, and the target needs in terms of changes to the current EDM system interface). The preparatory work resulted in the initiation of two project threads:

• The testing and implementation of optimal algorithms to allow autonomous extraction of selected accounting data from invoice images.
• Modification of the current EDM system interface to enable the automatic filling of extracted accounting data.

A genuine challenge was employing maximally efficient algorithms from the field of machine learning. These algorithms needed to have the capability to autonomously adapt to new patterns and invoice templates, and to extract desired accounting data (such as the document issuance date, sales date, payment date, contractor name, net and gross amounts, VAT rates, currency type, bank account number, and many others).

Results

The effectiveness of the proprietary solution implemented by FINTURE was validated through two-month testing periods of algorithms in the target environment (under expert supervision from designated client employees). The anticipated effectiveness of the INVOICE Reader module exceeded the base requirements of the client, achieving a stable accuracy level of over 85% for all processed documents and their extracted accounting fields.