In 1925 Wüstenrot brought building and home loan savings to Austria. Since that time, 440,000 homes were purchased in Austria with the help of Wüstenrot financing. Today the Wüstenrot Group is a Central European financial corporation. Around 2,700 employees successfully serve and support more than 2.2 million customers in Austria, Croatia and Slovakia, providing total solutions from a single source in the areas of savings plans, financing, pension provision and insurance.
Wüstenrot has been honored with awards in these areas and occupies first place in the OMG bank comparison 2013 confidence ranking. Through targeted customer and service orientation and the ongoing development of innovative products, Wüstenrot proudly holds the continuously growing trust and loyalty of its customers.
The Wüstenrot Group receives 25,000 data inputs every single day. The amount and types of input channels as well as the numerous kinds of documents (on paper, in e-mails, attachments, via the web) are increasing and it is becoming a herculean effort for the mail room staff to process the incoming information quickly and then pass it on to the correct department.
Increasing efficiency and quality means that even more structured data for each department needs to be collected, processed and validated. The logical answer was to standardize and automate the handling process for classification and data extraction from digital and analog information inputs.
After the evaluation of Proof of Concept (PoC) with several bidders, Wüstenrot chose Mindbreeze InSpire to automate their incoming mail classification. As early as the initial trial runs Mindbreeze achieved an accuracy rate of over 80 %.
To realize this level of accuracy, Mindbreeze was “fed” pre-classified documents in advance, teaching the system how to analyze composition, structure, words, word groups or text passages and link them to specific document types (building loan contracts, auto insurance, etc.). The analysis approach using new technology from the field of enterprise search (linguistics, semantics, stemming, etc.) combined with unparalleled recognition accuracy places Mindbreeze in a field of its own compared with all other providers.
Mindbreeze InSpire understands over 500 different types of file formats, enabling the system to easily classify and extract data from information collected from e-mails, Word documents, PDF files, and a plethora of other sources. Mindbreeze extracts the information directly from the original format. Thus, structured information is gathered not only from pre-structured content, such as the send-date of an e-mail, but also from unstructured texts.
Using pre-classified documents, the “training” of the Mindbreeze InSpire appliance began immediately after delivery. The preinstalled appliance with its self-learning system allowed for a rapid system start-up.
Today, documents are automatically OCR-scanned, categorized, enriched with metadata and reviewed. In the review, examining the content of the meta-data and calculating the recognition accuracy of each type of document are of crucial importance. If errors are detected, or if the recognition accuracy speed is deemed too low, the staff can check via the Fabasoft Capture client and adjust when necessary. Manually corrected documents are handed over to Mindbreeze for training. This automatically increases the recognition accuracy over time.
Since October 2014, Wüstenrot has been able to process information from multiple input channels, and has reaped the benefits of automatic classification and data extraction. The service quality and the mailroom throughput were increased. After achieving a certain accuracy level, incoming mail could be routed directly to the relevant department without manual inspection.
The previous solution for paper using form recognition has been replaced by future-proof technology. Wüstenrot now enjoys complete form freedom and the flexibility to meet the challenges of the future.
- 25.000 input entries per day
- Self-learning system with about 60,000 training documents
- Accuracy rate of over 85 %
- Multiple input channels
- Data extraction of more than 20 metadata in 138 categories of documents