Data extraction with AI at Bank11
Automated extraction of vehicle registration data using artificial intelligence
One of the business cases at Bank11 involved the daily and repetitive task of manually transcribing data from vehicle documents. In total, the employees processed around 190,000 vehicle registration documents per year, all done by hand. This process was not only time-consuming but also expensive and prone to errors. Moreover, it diverted employees from more critical tasks, such as providing excellent service and quickly processing credit approvals.
The goal of Bank11 was to automate these processes. While there are standard software solutions for document scanning, extracting data from vehicle registration documents comes with its own set of challenges. Therefore, a custom solution had to be developed.
// Application of Deep Learning OCR Tools
// Complex preprocessing tailored to vehicle registration documents
// Development of a proprietary validation process
// OpenCV's EAST
// Google Vision API
// Python
// Neural Networks
// Kanban
Our solution
As part of a proof of concept, we extracted the Vehicle Identification Number (VIN) from vehicle registration documents. In addition to reliably identifying the location of the VIN in the document, the varying characteristics of individual vehicle papers posed a particular challenge: inconsistent fonts and colors, different print intensities, prints over the form lines, and documents scanned upside down had to be cleaned up. This was done in the so-called "Scan Preprocessing". Using automatic image processing processes, the scans became clearer and more legible, allowing for standardized capture. The result: The VIN has been read with an accuracy of 92.31%. Before the use of our tools, the success rate was below 20% and, therefore, impractical in practice.
In the future, all data from vehicle registration documents should be extracted to avoid labor-intensive and error-prone manual entries, save costs and time, and relieve employees. The data from vehicle papers can also be used for further innovative optimizations. For example, customers could automatically receive information and credit approvals for a vehicle on their device via a smartphone scan. This would be a clear improvement in customer experience. In addition to automated workflows, the captured data can be processed profitably as a basis for further analysis purposes. With creative and innovative approaches and the ongoing digitization, Bank11 can potentially tap into even more potential through this form of data analysis and processing.
As part of our proof of concept, we worked with Qvest Digital AG on the goal of automating the way we capture vehicle registration documents (authorization certificates) and enhancing data quality. The result? We’ve been able to reduce processing time from one to two minutes to 30 seconds and the error rate from around 20% to 5% – with a total volume of approximately 100,000 letters. All in all, the colleagues from tarent implemented the project very quickly and professionally – and showed real passion for their work.
It was truly great to work on this exciting OCR project for Bank11. A fantastic challenge that we successfully addressed with creativity and the latest technologies. We also appreciated the good and straightforward collaboration with Bank11.
About our customer
Bank11 is a specialized financial institution in retail financing, providing competitive mobility and insurance offerings, primarily supporting medium-sized automotive dealers. Bank11 offers simple and affordably calculated financing products for end customers and robust liquidity solutions for automotive dealers. In the field of purchasing finance, Bank11, thanks to digital floor checks, provides significant efficiency gains, allowing for time savings and risk reduction. Competent and personal support through a specially trained field and internal team completes the service spectrum of Bank11.
