Claim adjuster notes, diary notes, and other forms of unstructured text data are largely untapped sources of detailed information on property-casualty insurance claims. These data can provide useful information on the particular circumstances of an incident, assignment of liability, and other details of a claim that may not be tracked in standard structured-data fields. For example, while the number of fields in structured data is usually limited to reporting information on one or a few causes of incident, mining unstructured data can be used to capture numerous causes of an incident. Furthermore, although there is often an initial assignment of liability, the assignment (or apportionment) can change as additional information becomes available. Claim adjuster notes and other unstructured data can be used to obtain current information on liability on an as-available, on-going basis.
Using real-world data on automobile complaints and defects, this session will describe procedures for extracting and compiling text data into a form useable for data analytics. The session will then illustrate how information gathered from text data can be incorporated into a predictive analytics model. This illustration will demonstrate with real-world data how information from text data can improve the fit of a model to data, and identify useful explanatory variables not otherwise captured in structured data.

Moderator/Panelist:
Philip S. Borba, Principal and Senior Consultant, Milliman, Inc.