Better Insuring With Information

July 31, 2000 | Last updated on October 1, 2024
3 min read

There’s an old saying “what you don’t know can’t hurt you.” But, in the insurance business today, this adage no longer holds true–what you don’t know can definitely hurt your organization’s bottom-line.

The danger comes in not knowing some of the more important information for strategic and successful decision making. For instance, which policyholders pose the greatest risk? Who is most likely to submit a fraudulent claim? Which customers are likely to defect to the competition? Without this information, insurance companies leave themselves exposed to high expenses and ever dwindling profits. This challenge lies at the heart of data mining technology.

At its most fundamental level, data mining is the process of selecting, exploring, and modeling large amounts of data to uncover previously unknown patterns. Insurance companies can use data mining to break down data groups based on specific characteristics, isolating key variables and performing forecasting and trending reports. By applying data mining techniques, insurance companies can:

Determine risk and establish policy rates;

Detect fraudulent claims; and

Curb customer attrition.

Critical to an insurance company’s success is the ability to set appropriate pricing rates. To do so, actuaries must be able to accurately determine a policyholder’s level of risk. Determining risk with a high degree of precision is, however, dependent on the interaction of several variables, interactions that can be overlooked without the aid of sophisticated data mining techniques. Data mining often improves predictive accuracy by segmenting data into more homogeneous groups. The data in each group can then be analyzed with high degree of precision.

For example, one insurance company found a segment of 18 to 20 year old male drivers had a noticeably lower accident rate than the entire group of 18 to 20 year old males. What variable did this subgroup share that could explain the difference? Data mining revealed that members of the segment in question had customized vintage cars which appeared to cause them to be more cautious in their use.

Fraud costs the Canadian insurance industry more than $1.3 billion every year. It stands to reason that insurers who are in a better position to detect fraudulent claims are also better able to offer competitively priced products and to maintain long-term profitability. So, is fraud a random phenomena, or is there a pattern companies can understand to protect themselves?

Fraudulent claims are typically not the biggest claims, because perpetrators are well aware that larger claims are scrutinized rigorously. As result, in searching for fraudulent claims, analysts must look for unusual associations or unexpected patterns in the data. Data mining techniques adept at finding such subtleties are cluster analysis, affinity analysis, decision trees, and neural networks. By comparing the expected with the actual, large deviations can be identified for further investigation.

For example, a US health insurance company used a database to compare a doctor’s claims against a larger historical base of data. The analysis identified several areas where claims exceed the norm. The insurance firm investigated and confirmed that a physician was submitting false bills. As a result, the doctor was forced to pay restitution and fines. Data mining saved this health insurance company as much as $4 million.

Another area where data mining can offer insurance companies enhanced performance and ultimately bottom-line profitability is the area of marketing. By drilling down into customer information through sophisticated data mining techniques, insurers can identify the characteristics of their most valuable customers and product lines for use in strategic marketing campaigns. This customer-centric approach can help insurance firms retain customers, acquire new ones and develop successful new product offerings.

For example, data mining can identify that a customer holding two policies with the same insurance company is much more likely to renew than a customer with a single policy. Similarly, data mining can discover that customers holding three policies may be more likely to switch insurance companies. For each group of customer, the marketing team can create targeted campaigns that provide an incentive to renew and a reward for customer loyalty. This targeted approach can reduce marketing costs by making campaigns smarter and more effective.

The Wired World welcomes your feedback. Contact us, via E-mail at <vspencer@corporate.southam.ca”>b>vspencer@corporate.southam.ca