Skip to main content
Data Mining in Insurance

Interesting article in Business 2.0 about Allstate's use of Data Mining.  What is not brought out in the article is that in order to get value from this data mining, Allstate had to use it to improve a transactional decision - the underwriting process. This means that a system, in this case the one being used by an agent, had to have the insight gained from the data mining embedded in it. Now clearly if the person in the article had gone online or called a call center you would have wanted him to get the same offer.

All of this is the basic premise behind EDM - automate decisions and use data analysis to improve them.

It should be noted, in case you are worried, that credit is a strong individual predictor but when combined with other predictive variables provides great underwriting and pricing segmentation. Knowing how to identify these variables and account for their relationship with other variables is how you really get a return from using credit data - you can find good overall risks with poor credit and poor overall risks with good credit. Fair Isaac experts on this topic tell me that on average credit accounts for about 15% of the value of a risk model with other factors accounting for the other 85%.

It is also true that anyone wanting to use credit as described in the article must also make sure that their use of credit interacts with the vast set of rules that exist in the underwriting process. There are eligibility rules, knock-out and referral rules. Managing the interaction of models with rules is key to an improved business result - this is a key premise in EDM.

Hundreds of insurance companies use this kind of information thanks to an offering from Fair Isaac called Credit Based Insurance Scores - there's loads more information at These allow insurers who can't afford the resources described in the article to compete with  large companies who can do this for themselves.

Secondly several states have recently begun restricting the use of credit data so anyone doing this had better have some agility built into their process to respond to changes in regulations state by state - a business rules management system would seem to be called for - as well as a partner that understands the use of credit in risk models.

Thanks to my colleagues Wendell Larson and Lamont Boyd for their expertise on this one.

related posts