In spite of centuries of sales experience, most insurers rely heavily on price for acquiring and retaining customers. The reality is that customers base insurance purchasing decisions on many factors. While this does not mean that we can ignore pricing, it does mean that alternatives exist for insurers to remain competitive.For some companies, the frustrating cycle of hard and soft markets will not exist. By relying on the power of data mining, they can maintain the consistency and accuracy of their underwriting decisions; they can significantly reduce the impact of fraudulent claims; and they can have a better understanding of their customers' wants and needs.
Data mining is the exploration and analysis of potentially large volumes of data to discover or confirm useful relationships, patterns and rules. These discoveries are then used to make some prediction or inference about future behavior.
Two classes of analysis
Within the field of data mining, there are two broad classes of analysis: supervised and unsupervised learning.
Supervised learning uses predictive mathematical modeling techniques against a database populated with known outcomes.
For example, the database may be a list of policyholders' loss histories. The known outcome, or target, would be whether policyholders have had a loss during a specific time period or the magnitude of a loss.
Unsupervised learning relies on data with no defined target-using tests of association, segmentation, clustering, and advanced forms of compression neural network modeling techniques.
This exploratory analysis can identify policy endorsements most often purchased together so new product bundles can be created to better suit the customer needs.
The majority of personal-lines insurers use credit scores in their underwriting decisions. But if most insurers use credit scores in the same way, how can one gain any competitive advantage over another?
The answer: build your own model. By using your own data, you can create scoring models that have greater predictive capability than credit scores alone. When combined with your company's billing experience, aggregated loss history and scores of other characteristics, a custom scoring model can ensure that your underwriting decisions are correct.
The technology exists for building reliable predictive models rooted in statistics that consider many more factors than an underwriter can.
Insurers can also use data mining to improve detecting and predicting claims fraud.
Typically, insurers rely on database searches of past losses, which depend on the adjuster correctly identifying fraud, and referring cases to the special investigations units (SIU).
Most insurers also use a rules-based approach to "red flag" claims. As is the case for underwriting, there is generally no statistical basis for the rules. Fraud rules are based on limited past experience and are ineffective in detecting emerging fraud techniques.
Data mining improves fraud detection. For example, to uncover anomalous claims behavior that may indicate fraud, unsupervised learning is an excellent approach.
The use of compression neural networks to model claims behaviors produces models that function as anomaly detectors. By not requiring a database of known frauds, this system creates models whose interpretations can be tuned to minimize false positives and maximize the proper use of the SIU's resources.
Because relatively few fraudulent claims are detected-and even fewer are prosecuted-under current business practices, the database of known outcomes required for supervised learning is problematic.
First, the data contains numerous false negatives. A predictive model built using incorrectly classified data would be likely to ignore certain types of fraud and would be less able to identify new fraudulent behaviors.
Reliable results
A combination of techniques, beginning with unsupervised learning and using its findings to build a more accurate database, would yield much more reliable results. Once the accuracy of the database's classifications is established, supervised techniques can be used to improve the detection process.
Finally, staff underwriters, marketers and actuaries can now build predictive models that identify policyholders most likely to depart, new customers most likely to respond to product offers, and existing customers most inclined to purchase additional products.
Data mining enables insurers to help their agents, both captive and independent, simply by giving them pre-qualified leads. By making the right offer to the right customer at the right time, the company adds value to the relationship.
Responding to competition with lower prices alone has never been a long-term solution. With data mining, insurers have many ways to improve their competitive position: better underwriting, improved fraud detection and more effective marketing campaigns. How you choose to compete is up to you.
David West is the global insurance strategist at SAS Institute Inc.
