Algorithm improvements and bugfixes
Multiple improvements for the DATEV integration. Improved document handling with different booking names and better error handling.
To increase accuracy of data extraction, we have changed our deep learning model architecture and retrained millions of invoices. Results are higher accuracies for data extraction for the following fields and countries, shown in the graphic below. Accuracy increased by 2% on average for German invoices, 5% on average for Non-German invoices (French, English, Spanish).