INVOICE PROCESSING

Institutions receive masses of invoices daily. These sparse documents come in a variety of inconsistent formats. We generally believe that robotic process automation (RPA) is a dead-end street, invoices are the most obvious example of its failure.

Successful processing of invoices can only be done through artificial intelligence.

Combining several models, we are achieve superior results in automatically processing invoices as well as receipts. Our service extracts all common data items, such as vendor, date and invoice number, and it also extracts the description of the provided services or products, creating entities which can be used to cross reference with products and services from other documents.

THE AUTOMATED INVOICE PROCESSING STEPS

1

SECTION CLASSIFICATION

Invoices are mostly documents with sparse text. To be able to process them correctly, a classification of sections is essential. For example, differentiation between sender and receiver is often only possible through graphical interpretation. We have trained a dedicated Convolutional Neural Network with more than one million annotated invoices which results in an unparalleled accuracy.

2

OPTICAL CHARACTER RECOGNITION (OCR)

After graphical interpretation, we transform images to text using an LSTM model of tesseract, enhanced by domain-specific training data. Tesseract has good accuracy out of the box but enhanced with a financial library we have increased the accuracy considerably.

3

NAMED ENTITY RECOGNIZER

Products and services on invoices need to be recognized, not only for that particular invoice but across invoices and also for matching with other documents such as offers and proposals for products and services. Building on top of MUC7 we have added annotations for over one million finance domain specific invoices which improves results significantly.

Operating in the financial industry, we are aware of the ubiquitous regulatory, compliance and data protection matters. Keeping these in mind, we have built our service to be deployable within the client's infrastructure or in a secure cloud infrastructure. We are currently advancing our models to continuously improve using reinforced learning.

INTERESTED TO LEARN MORE?

Drop us a note and we will get in touch with you.

Please enter your name!
Please provide a valid email address!
Please give us some details so we can inform you better!