It is no secret that the Insurance industry is one of the biggest employers of data scientists and analysts. This report by Accenture talking about the global talent crunch in Insurance shows a shortfall of 15,000 jobs in 2015. Within the Insurance sector, the underwriting & rate setting is the most critical analytics function. If you quote an insurance rate that is too high you lose market share and if you set it too cheap, you have underwriting losses. This fine line has meant an extremely complicated analytical model driven by hundreds of variables, all trying to figure out what your resulting claims would be and then setting an insurance premium from that.
In addition, companies & insurance agents use “art” on top of the science to set the final rate. The “art” part of the equation includes how much cross sell & up sell opportunities there could be with other products in the future. For example, if there is a probability for you to purchase car insurance in the future, this implies that the cost of customer acquisition will reduce on the entire bundle which will in turn make the customer more profitable. This “Art” is normally left upto the insurance agent’ judgments when the final quote is done. Some agents can even take hits on their commission depending on the “right” customer.
vHomeInsurance (www.vhomeinsurance.com), a home insurance analysis service, shows an illustrative “Art” Vs “Analytics” example for a typical customer looking for Homeowners Insurance in Dallas. A standalone home insurance rate in Dallas could be anywhere between $850 to $1200 depending on a number of variables such as your street address, age, home value, number of bedrooms year your house was built. At the same time, if the customer wanted both auto & home, they could get anywhere between a 10 to 30% discount saving a few hundred dollars in the process. This discount is offered because of lower customer acquisition & customer service costs from a bundled home & auto service.
But- what if the Dallas Homeowner is not sure of the auto & home bundling and wants the quote only for home insurance. The insurance agent who knows the background of the customer may still give some discount for a “potential” future bundling. This estimation of “potential” is what leads to a lack of discipline and final underwriting losses for the insurer. Warren Buffett repeatedly coaches his insurance executives that the art of saying no is more important than saying yes to have a sustainable insurance business. However, on the other hand, the potential for future bundling is real.
Realizing the potential in a disciplined & analytical manner would involve building an aggregate variable called “Probability of Cross Sell” BEFORE a person becomes a customer. This implies that our probability will need to be developed without any past transactional behavior of the potential customer. The model would need to be calibrated just on 3 Factors 1) What the customer has provided 2) External Data 3) Internal Data. Merging all these activities into a single unified view for the front end agents to quote a customer is not easy. But, if one could get it right, there could be significant benefits for the insurer & the customer in moving from art to analytics.
vHomeInsurance (www.vhomeinsurance.com ) is a home insurance research & analytics service that focuses on the micro factors such as location, house & neighborhood as well as big picture technology , consumer and analytics trends that impact your homeowner insurance.
Try deep learning using MATLAB