Invoice Recognition: Record Line Items with OCR

For a long time, invoice recognition was primarily available for invoice details such as date, number, creditor, debtor, total amount, etc. However, the development of artificial intelligence has increasingly enabled machine learning systems. At BLU DELTA it is part of our culture that we regularly read research papers for OCR and document capturing in detail in order to then implement the latest technologies such as deep learning and transformer models. With this technology, it is possible to extract more and more specific information from documents with ever greater accuracy. Therefore, automatic invoice recognition at line level (= line items) is also possible with the BLU DELTA OCR API and invoice processing!

Why do you need line item data in accounting?

The automatic extraction of the head (or header) information of an invoice is of course the most important thing to convert many invoices into an automated booking proposal. The iOCR and OCR technology has already proven itself in this area and has led to an increased recognition rate for invoices. With standardized ledgers, the automation of posting proposals is getting better and better. However, it is not always possible to create a complete posting proposal with the header information of a receipt. Sometimes a so-called “splitting” of an invoice is necessary. So not every invoice line (or line item) has to be posted to the same account, cost center or project. The basis for an automated split posting can be laid here by identifying or recognizing individual line items on invoices.

How does BLU DELTA recognize information on an invoice or document?

Artificial intelligence can be trained. A large number of example receipts and invoices are “shown” to an intelligence. At BLU DELTA we prepare such examples. Based on input from all our customers, we prepare the information for the training, marking date, number, amounts and also position data. We ensure the accuracy of the sample data. The artificial intelligence is then trained with many example calculations and the quality is measured with a second set of example documents. The technologies behind it are deep learning and transformer architectures.

Display of found header and position information

Any invoicing program can handle PDF. That’s why we at BLU DELTA generate a second PDF in addition to the original, which includes the desired information such as total amount, tax amounts, order number, delivery date, item numbers, etc. Thus, the visualization can be accomplished in any standard software.

Do you want to automate or improve your accounting?

If so, you will soon notice that the high-quality and automated recognition of the invoice is the key to all follow-up processes. Practically all subsequent decisions depend on the recognized data and line items. The recognition should deliver good results without a great deal of manual effort with a high and/or growing number of suppliers at the same time. Only certain technologies are well suited for this. Doesn’t your system do that? Then they should question their technology in detail and assess the underlying concepts for their individual situation. At BLU DELTA we are always ready for an open discussion. We are happy to take an hour for a free consultation.

Note: You can find more details on how an OCR works and why it forms the basis for RPA and other automations in our article "What is OCR?".