Intelligent Document Analysis and JAVA-based accident claim Processing

at the Berufgenossenschaft für den Gartenbau
(Horticultural Trade Association)

Project phase I

Initial Situation		The Horticultural Trade Association (GBG) is the accident insurance institution for all businesses in the field of horticulture.
Requirements		The project's aim was to initially replace paper files in clerical accident processing (about 10,000 pages of incoming post per day) with electronic files. Clerical processing was to become quicker and more efficient through a shortened period of time between the receipt of incoming post and clerical processing, simultaneous accessing of a single file by several relevant clerks, and an improved integration into the system of information for the Agricultural Social Insurance (ISLSV).  At the same time, using modern document analysis technology, the incoming documents were to be classified according to their document type (accident report, doctor's report etc.), and data relevant to processing (such as accident numbers, invoice sums) were to be recognised automatically wherever possible, in order to tighten up the work procedure between postal receipts and clerical processing.  In addition, all outgoing documents, including host-based processed and MS Office documents, were to be archived in an electronic file.
Project Description		In order to make the most of BGB's existing infrastructure (an S/390 System) and to achieve a short reply period, the VisualInfo/390 archiving system was decided upon. Since only the contents and not the format of some documents differ (e.g. prescriptions and doctors' orders), the automatic recognition process could only be contents-based.  The product chosen for this was Paradatec GmbH's PROSAR-AIDA, which had already established itself as the market-leading solution for private health insurers. The capture system used was InputAccel, from the Action Point company (formerly Input Software), which controls the entire system of scanning, indexing and exporting images and index files to the archive system. InputAccel excels through its easy conifgurability and open interfaces, allowing smooth integration with PROSAR-AIDA.  GBG differentiates between a total of 700 document types and 25 document groups, which differ in terms of their fields and index criteria. Since the determination of the document type cannot be achieved at a rate of 100%, the indexing application must, if necessary, dynamically change the input interface when the indexer enters a document type. The assignment of index fields must thereby be dynamically selected from the archive system, in order to avoid erroneous double administration. Paradatec GmbH's ProKey capturing system is also employed here, as it can be configured via a DLL interface in a highly flexible way. The allocation of incoming documents to clerks' in-trays is carried out using the indexed accident numbers directly via the host-based ISLSV. The main underlying idea is to reduce software administration at work places in the company. The application is based on Java applets currently running on an NT server (set up for up to 100 users), and is automatically distributed from there. The only requirement is an internet browser (such as MS IE5) for the clerk's work place computers to participate in the process. The use of JAVA means that the clients are not even dependent upon one operating system. In terms of hardware, only ergonomic requirements (such as monitor-size and processing speed) need to be taken into account. besides the usual standard functions of a DMS, the functions necessary for processing accidents were embedded in the client. Thus, in addition to mailbox, search, memo and display functions, the client's operative host application can be integrated via a JAVA host emulation (HOBLink Jterm).
Benefits		The new electronic file allows for parallel work processes which hitherto had to take place consecutively since a paper file cannot be used by two clerks at once. With the new archive, the accident report is not only sent to the relevant accident clerk, but also is simultaneously sent to the technical supervisory service to register the statistics on the cause of the accident.  The document volume of 10,000 pages per day is processed by an IBM Netfinity multiprocessor machine directly after scanning, so that the analysis results (document type, accident number and, when relevant, invoice sum) is presented within minutes. Through recognition rates of 66% to 100% depending on the document type during document classification, the manual indexing effort can be kept to a minimum.  The entire system can be operated by the client, so that GBG can offer archiving or even document analysis as a service to smaller trade associations for whom the investment costs for their own EDMS solution do not seem warranted.
Publications		Lecture: Lecture at the AIIM International Conference / DMS EXPO 2000, Kurt Klemichen (GBG):Intelligent document analysis and JAVA-based accident claim processing Presentation sheets and handouts for this lecture are available from the lecturer.
Contacts		at the GBG: Kurt Klemichen, Telephone +49 561-928-2460 at IBM: Ingo Bahrs, Telephone +49 511-516-3121 at Paradatec: Bernt v. Nottbeck, Telephone +49 531-23815-0