X hits on this document





11 / 20

Transforming Underwriting

MARCH 2004

an expected loss of US$300. Carrier A is going to be insuring most of those applicants. This phenomenon, attracting many underpriced risks, is called adverse selection. Carrier A is incurring adverse selection. By pricing correctly, Carrier B has a distinct competitive advantage.

In recent years, some insurers have begun to use a simple but powerful tool for developing rates and making underwriting decisions: predictive scores. Predictive scores give an insurer a more transparent and predictable way to assign applicants to tiers, make underwriting decisions; and determine rates.

A predictive score is usually calculated from 8 to 20 variables. These variables could be such things as a house’s size in square feet and its year of construction, or the make and models of insured automobiles and traffic violations by drivers. Each variable must have either a distinct value (less than 1,500 square feet, 1,500 to 2,500 square feet, etc.) or a condition (there has, or has not, been a moving violation in the past three years). Weights are assigned to the variables, e.g., the most desirable square footage has twice the weight as the most desirable age of construction. The total of weighted values and conditions produces a predictive score, which is expressed on a maximum/minimum scale, for example, 200 to 800.

Insurers use analytics to select the variables; to assign points to values and conditions; and to determine weights. Analytics uses sophisticated methods of statistical analysis to discover relationships and predict outcomes. Analytics requires the use of various forms of software, such as data warehouses and business intelligence applications.

After analytics determines the method for creating predictive scores, the speed and flexibility of business rules engines make them the preferred way to implement and modify those predictive scores.

Business rules engines are also well suited for using predictive scores in the underwriting process. For example, an insurer can calculate a predictive score, then use other rules to place an application in a tier (e.g., does the applicant have other policies with us?) or to “knock out” an otherwise good application (e.g., is there a drunk driving conviction?).

Predictive scores are double-edged swords. If they accurately predict losses, they will mow down the competition. If not, they will cause self-inflicted wounds.


Predictive scores, and business rules-driven underwriting and rating decisions all require data. This data can originate externally from third parties (driving records, claims history with other insurers, credit scores, property inspections, etc.). For renewals, it can originate internally within the insurer (claims and payment history, etc.).

© 2004, Celent Communications. Authorized reproduction permitted.



Document info
Document views74
Page views74
Page last viewedMon Jan 23 02:33:05 UTC 2017