Automatic Analysis of incoming Mail at DKV

Project stage I: Indexing
Project period: September 1998 to April 1999

Original situation	About 18 000 pieces of client-mail with a volume of 110 000 pages of paper arrive daily at Deutsche Krankenversicherung AG (German Health Insurance plc) in Köln. At peak times this can increase to a daily volume of 150 000 pages per day. In processing, the entire incoming mail is initially scanned and manually indexed according to the following criteria: insurance-number (per letter), document-type and insured person (per page). On the basis of these criteria, the relevant employee can be determined and the document can be forwarded electronically. Further processing involves collecting a series of data manually, such as invoice amounts.
Requirements	In November 1997 DKV Deutsche Krankenversicherung Köln invited tenders for an automatic recognition system which could fulfil the following conditions:  smooth integration into DKV's computer systems complete analysis of all incoming mail. The result should acsertain the following information: Each side's document-typ. Recognition of several hundred different document types based on key words in different places in the text, along with typical pictoral information (logos), must be possible. The insurance number, which can appear anywhere on a file's page, or which can be determined by the address on the page. The insured person, who can be linked to a specific receipt. This requires an analysis in search of names in combination with an insurance number. Depending on the document type, the second stage of the project should determine further information. This includes:  In formatted texts (e.g. doctors' invoices), specific key information should be located on the basis of key words (such as "total sum") so that linked information (such as the amount of money invoiced) can be recorded. In documents containing tables with benefit data (GOÄ and GOZ tables), the entire contents of the table should be read irrespective of its layout, so that this information can be forwarded to the DKV host computer.  After  a first selection process, DKV invited 4 of the companies who handed in tenders to present the solutions they had offered by means of a prototype. At the end of April 1998, following extensive analysis of the presented systems, DKV decided to accept IBM Deutschland as the main contractor, along with its solution based on Paradatec GmbH's PROSAR-AIDA recognition system.
Project description	Under the leadership of IBM, project work began as early as May with an analysis of the status quo and the formulation of a detailed list of specifications which both described DKV's requirements for the recognition system's scope of implementation, and also precisely documented the necessary integration work. Based on this list of specifications, a project plan was developed which foresaw the completion of the project's first stage, including the automatic generation of indexing data (document type, insurance number, insured person) before the end of the year.  The list of specifications was approved by DKV in September 1998, whereupon the implementation order was made. The project's first stage began production as planned in December 1998. While the system was used to capacity during daily production, a four month optimising period ended in April 1999.  All promised recognition rates were not only achieved, but in most cases clearly exceeded. On the basis of the successful implementation of the project's first stage, the immediate conception of its second stage began.
Publications	Lectures: Lecture at the CeBit 1999 by Alf Jarosch IBM-user insurers' conference 1999, Peter Grunewald Presentation transparencies and/or handouts for these lectures are available from the lecturers. Newspaper reports: Frankfurter Allgemeine Zeitung vom 6.11.1999 (in German): "DKV will Einfluß auf Ausgaben vergrößern"
Contacts	at DKV: Jürgen May, telephone +49-(0)221-578-2864 at IBM: Ronald Fritz, telephone +49-(0)30-3115-1641 at Paradatec: Bernt v. Nottbeck, telephone +49-(0)531-23815-0