I saw this article in National Underwriter - Data Use, An Edge For Some Insurers: Consultant. In this the authors of the study say
"Property-casualty companies that reject traditional business models in favor of the creative use of data mining and client segmentation will have an advantage in the coming years"
This reminded me of the discussions I have with insurance customers around being adversely selected against and how decision management, especially decision management that takes advantage of data mining and predictive analytics, can help. So what is "being adversely selected against" I hear you say. Well consider two companies, one has three prices (for good risks, moderate risks and bad risks respectively) and the other has nine (very good, somewhat good, slightly good, very moderate, somewhat moderate etc). This means that if a specific customer is in the good tier for company 1 they might be in any one of three tiers for company 2. Now company 2 is smart - they make their price for their "very good" segment better than company 1's "good" segment, their "somewhat good" segment price about the same and their "slightly good" price a little worse. Let's consider three potential customers:
- Customer A is a very good risk and so gets a better price from company 2 (who puts them in their "very good" segment) than from company 1 (who puts them in "good")
- Customer B is a somewhat good risk and so gets about the same price from the two companies
- Customer C is a slightly good risk and so gets a worse price from company 2 than from company 1
Given this, customer A is more likely to pick company 2, customer B might go either way and customer C will likely pick company 1. Repeat over many customers and company 1 is likely to end up with more customers like C and fewer customers like A. Instead, therefore, of having customers spread evenly across the segment company 1 will have more customers at the low end of the range ("somewhat good") than at the high end ("very good"). This in turn means that the average customer for company 1 will be somewhere between "somewhat good" and "slightly good". In contrast, company 2 will get more "very good" customers and an average between "somewhat good" and "very good". For company 1 this means they are being adversely selected against - they have priced their segment based on the average but are getting people who average worse than that. If your competitors have more fine-grained pricing than you do, this is what happens. You end up with a customer portfolio that tends to be worse, on average, than you expect.
How do you avoid this? Well by more finely targeting and segmenting your customers. This takes analytics or data mining. If you can use your data to better predict risk and then combined this risk assessment with the rest of your data before segmenting your customers then you can get more fine grained segments. The two graphics show how more segments help you better match the reality of a continously changing risk curve. There's more on segmentation here and on the insurance industry and EDM here.
The key, as the study's author said, is to take informed risks.