The problem is that many companies rely too much on the model and not enough on skilled/tenured underwriters…particularly in Commercial P/C. The models should be used as a tool and not the whole pricing decision.
Underwriters cannot predict human behavior accurately. The question is who can get closer. More sophisticated models will eventually outperform underwriters at a lower cost.
This predictive modeling process or as our company calls it “expert underwriting” has been a farce. Many long term households that are profitable are being canceled without any reasoning taken into consideration by underwriters for facts that prove otherwise than the model they are using suggests. Once the insureds name comes up in the model it’s as if their on a death list without any hope of reprieve even though as in one such case I compiled the numbers and the household was $25,000 profitable overall from 15 years of longevity. Where is the common sense in this process? If the bottom line is improving profitability then why do these models resulting in accomplishing the opposite and canceling profitable households and yet underwriting won’t cancel a household that we request to be canceled due to bad claims loss history. This predictive modeling is a dud that has blown up in the face of our insurance company and caused many more problems than any benefits or profit it may have created. Bad system in my opinion that doesn’t accomplish what it is intended to do.
Shawn – by trying to calculate profit on one single homeowner exposure you are showing that you do not understand personal lines pricing. If they have a total house fire tomorrow, then this risk will be “unprofitable,” right? That is not the correct or fair way to view things. In PL, you have to use the good ole law of large numbers and predictive analytics is an innovative way to do so. I agree with all that is said about using common sense, and knowing when to deviate from the model, but the use of predictive analytics in PL is a great leap forward in helping us to understand which exposures present the greatest likelihood of future losses. As for Commercial Lines, I can’t speak to that. CL seems to be a whole different ballgame than PL.
I’ve yet to see the predictive model used by one of the top Carriers that predicted the 2008 financial crisis and what it would do to E&O and D&O policies. Some Carriers are paying for that dearly right now.
A predictive model is only as good as the person programming said model.
Travelers and Hartford are already using them. This is the begining of the end for Marketing Reps and Underwiters that write BOP’s, CA, Property, etc. Sad but already happening with the big boys.
Predictive modeling works extremely well with large homogeneous groups buying a genetic insurance product in an environment where accurate risk and peril information, at point of sale, is available. It is why GEICO and Progressive have risen to the top ranks of auto insurers. Other lines of business will follow the saturation level of auto predictive modeling, when reliable information sources are developed for application in real time processing. Until then underwriting jobs are safe(r). Looming in the background is “disparate impact” which could be the fly in the ointment for companies that rely heavily on predictive modeling in personal insurance.
As a customer, how do I avoid the outrageous premium increases being heaped upon me by Farmers? My premiums have gone up 25% two years in a row with no record of any claims. Their comment, we need to protect ourselves from what is about to happen.
Anybody out there want a new customer?
The problem is that many companies rely too much on the model and not enough on skilled/tenured underwriters…particularly in Commercial P/C. The models should be used as a tool and not the whole pricing decision.
Predictive models will NEVER be able to predict human behavior and therein lies the problem.
Underwriters cannot predict human behavior accurately. The question is who can get closer. More sophisticated models will eventually outperform underwriters at a lower cost.
This predictive modeling process or as our company calls it “expert underwriting” has been a farce. Many long term households that are profitable are being canceled without any reasoning taken into consideration by underwriters for facts that prove otherwise than the model they are using suggests. Once the insureds name comes up in the model it’s as if their on a death list without any hope of reprieve even though as in one such case I compiled the numbers and the household was $25,000 profitable overall from 15 years of longevity. Where is the common sense in this process? If the bottom line is improving profitability then why do these models resulting in accomplishing the opposite and canceling profitable households and yet underwriting won’t cancel a household that we request to be canceled due to bad claims loss history. This predictive modeling is a dud that has blown up in the face of our insurance company and caused many more problems than any benefits or profit it may have created. Bad system in my opinion that doesn’t accomplish what it is intended to do.
Yes, no question that the models should be a tool and not the exact pricing tool that they have become.
Shawn, so 15 years is a long enough window of time to accurately assess storm potential?
Shawn – by trying to calculate profit on one single homeowner exposure you are showing that you do not understand personal lines pricing. If they have a total house fire tomorrow, then this risk will be “unprofitable,” right? That is not the correct or fair way to view things. In PL, you have to use the good ole law of large numbers and predictive analytics is an innovative way to do so. I agree with all that is said about using common sense, and knowing when to deviate from the model, but the use of predictive analytics in PL is a great leap forward in helping us to understand which exposures present the greatest likelihood of future losses. As for Commercial Lines, I can’t speak to that. CL seems to be a whole different ballgame than PL.
I’ve yet to see the predictive model used by one of the top Carriers that predicted the 2008 financial crisis and what it would do to E&O and D&O policies. Some Carriers are paying for that dearly right now.
A predictive model is only as good as the person programming said model.
Travelers and Hartford are already using them. This is the begining of the end for Marketing Reps and Underwiters that write BOP’s, CA, Property, etc. Sad but already happening with the big boys.
Predictive modeling works extremely well with large homogeneous groups buying a genetic insurance product in an environment where accurate risk and peril information, at point of sale, is available. It is why GEICO and Progressive have risen to the top ranks of auto insurers. Other lines of business will follow the saturation level of auto predictive modeling, when reliable information sources are developed for application in real time processing. Until then underwriting jobs are safe(r). Looming in the background is “disparate impact” which could be the fly in the ointment for companies that rely heavily on predictive modeling in personal insurance.
As a customer, how do I avoid the outrageous premium increases being heaped upon me by Farmers? My premiums have gone up 25% two years in a row with no record of any claims. Their comment, we need to protect ourselves from what is about to happen.
Anybody out there want a new customer?