The client is one of the leading providers of consumer loans and leasing for private individuals and small businesses in Switzerland. The bank employs approximately 300 people, with customer loans amounting to over CHF 2 billion.
Data extraction from payrolls
Over 40'000 documents per year
Head Risk Processes & Framework
To apply for a loan, clients must submit various documents, including several pay slips for verification. The bank receives over 10,000 documents a year. The specialists manually record various predefined data points from these documents, check them, and determine a credit score based on the information obtained. Given the large variety of payroll documents with different layouts and structures, a static, template-based approach was out of the question in the past.
With the support of intelligent document processing (IDP) expert, Acodis, a new attempt was made to automate the process with the aim of introducing a process-supporting platform for automated payroll data extraction.
After setting up a representative training set of 1,300 payslips, a deep learning system was trained and evaluated. The training set included documents from hundreds of employees with more than 40 relevant data points. With an F1 score of 0.93, the resulting machine-based system achieved practically the same results as the reference values from humans (0.94) at a massively higher speed.
After these pleasing results, the platform was made available to other employees via a smart web application. They can now initiate the readout process via an upload (API or via the web browser) and the processing steps (OCR, classification, extraction, normalisations) are carried out automatically. Via a graphical user interface, employees can visually carry out cross-checks and make any necessary adjustments. Afterward, the structured information is automatically transferred for credit analysis and a dossier is created and filed for each customer.
The Acodis platform comes with the following key features:
With OCR, data extraction & validation
Easy to use and allows a fast labeling process
For quick visual inspection
For accurate analyses
In order to automate and make adjustments
To speed up internal processes
Thanks to intelligent document processing (IDP), the customer benefits from the following added values, among others:
With Acodis IDP, the customer has a modern application for the automated capture of payroll data - anytime, anywhere. The solution is flexible and allows document analysis to be seamlessly integrated into existing processes and to include other use cases such as fraud detection.