Demonetizing Insurance

[EricSchleien]

http://www.diamandis.com/blog/demonetizing-insurance

[rkbabang]

As prediction models get better, insurance rates will go way down for those least likely to make a claim and way up for those most likely.  Followed to its logical conclusion as prediction gets better and better eventually insurance prices will be approximately the same as the expected costs to you if you did not have insurance.  People will only be able to insure against things that are completely random and/or very rare.

 

[rb]

I think that the actuarial part of insurance (prediction models) is actually pretty good.

 

The way I see it the black mark on insurance is the sales and distribution model, the broker networks specifically. Most retail insurance policies are generic commodities. I can't think of any other commodity product where its sales person gets 15-20% of the price to sell it. Switching to a direct GEICO like model would increase efficiency and decrease premiums but insurance cos have been very reluctant to ditch brokers.

 

[rukawa]

I don't think better prediction will make a huge difference. Insurance companies have had hundreds of years to get this right and I would guess that all the biggest factors they already know about. And many of the rest they can't get info on because of privacy laws. 

 

As for fraud, computers will never be able to help with that. Insurance is the one place where you need real street smarts and cynicism. The human element is tremendously important. Everybody is a shyster in insurance. The people who make claims, the brokers, the service providers, everyone. Peer-to-peer bullshit and digital masturbation is most likely to result in a lot of bankrupt startups on a scale that will make peer-to-peer lending look sane.

[Jurgis]

As for fraud, computers will never be able to help with that.

 

Not true. Every organization that deals with fraud already runs everything through automated fraud detection. And there's a lot of money/effort spent on reducing both false negatives and false positives. There are humans in the loop for some of the fraud detected (although your CC gets locked without human in the loop likely), but you said "help" and there's definitely huge computer help in flagging/filtering already.

 

Although I agree that peer-to-peer startups are likely either go BK or not pay out claims in 3 minutes as advertised.

 

 

[rkbabang]

I don't think better prediction will make a huge difference. Insurance companies have had hundreds of years to get this right and I would guess that all the biggest factors they already know about. And many of the rest they can't get info on because of privacy laws. 

 

As for fraud, computers will never be able to help with that. Insurance is the one place where you need real street smarts and cynicism. The human element is tremendously important. Everybody is a shyster in insurance. The people who make claims, the brokers, the service providers, everyone. Peer-to-peer bullshit and digital masturbation is most likely to result in a lot of bankrupt startups on a scale that will make peer-to-peer lending look sane.

 

Insurance companies have had hundreds of years to get this right for large group of people.  I think the difference is that the sensors will for the first time allow insurance companies to get it right for the individual.  Which is moving away from spreading the risk and no longer allowing those who engage in risky behavior to free load on those who do not.  Obviously my statement about following it to its logical conclusion is not going to happen in real life, because prediction will never be 100% accurate. But the better insurance companies get at predicting your risk as an individual, the more your individual rates will reflect that risk and the closer you get to that ideal case I mentioned.  You can think of insurance rates based on perfect knowledge as a target the way the market approaches the evenly rotating economy.  i.e. it never gets there and real world changes in peoples' actions and mother natures' whims move the target continuously.  Obviously the better sensors and real time AI predicts every individuals actual risk, the closer the rates will get to that target before it moves.  The difference from the past is the focus on each individual in the calculations rather than on large groups.

 

 

[glorysk87]

I would think there's a pretty obvious point missing from this thread. That point being that no matter how good predictions get, part of the determinant for pricing is what people are willing to pay.  And seeing as how insurance has a pretty large psychological aspect (people fear being caught unprepared by the unknown - car crash, death, house fire, etc), there is a level of irrationality that causes people to pay up for insurance.

 

So one would think that no matter how efficient the predictive models get, insurance companies should be able to build margin into their products. Unless you think that people will suddenly and uniformly become completely rational.

[rb]

I would think there's a pretty obvious point missing from this thread. That point being that no matter how good predictions get, part of the determinant for pricing is what people are willing to pay.  And seeing as how insurance has a pretty large psychological aspect (people fear being caught unprepared by the unknown - car crash, death, house fire, etc), there is a level of irrationality that causes people to pay up for insurance.

 

So one would think that no matter how efficient the predictive models get, insurance companies should be able to build margin into their products. Unless you think that people will suddenly and uniformly become completely rational.

Yea, this is not right. No one is saying that insurance companies won't be able to build in margins but they won't be very big. Since insurance is (mostly) a commodity product and there are (fairly) low barriers to entry insurance cos don't have much pricing power. Basically the pricing is driven by the costs. Mainly claims, sales, fraud, and admin. And no, the consumer is not willing to pay up for insurance. The consumer wants the lowest possible price for insurance.

 

I'm also not sure how motivated the insurance companies are to make make this big investment in technology for sensors, etc.  If the tech improves pricing efficiency then the consumer gains/ looses by more predictive accuracy of risk but it does not increase the profitability of the pool for the underwriter so there's not much incentive to spend the money.

 

In addition if the current models are pretty good at predicting risk (I think they are) and there is marginal improvement in pricing accuracy -say 5-10% increase in accuracy - the cost of tech won't be worth it for underwriters.

[rkbabang]

I would think there's a pretty obvious point missing from this thread. That point being that no matter how good predictions get, part of the determinant for pricing is what people are willing to pay.  And seeing as how insurance has a pretty large psychological aspect (people fear being caught unprepared by the unknown - car crash, death, house fire, etc), there is a level of irrationality that causes people to pay up for insurance.

 

So one would think that no matter how efficient the predictive models get, insurance companies should be able to build margin into their products. Unless you think that people will suddenly and uniformly become completely rational.

Yea, this is not right. No one is saying that insurance companies won't be able to build in margins but they won't be very big. Since insurance is (mostly) a commodity product and there are (fairly) low barriers to entry insurance cos don't have much pricing power. Basically the pricing is driven by the costs. Mainly claims, sales, fraud, and admin. And no, the consumer is not willing to pay up for insurance. The consumer wants the lowest possible price for insurance.

 

I'm also not sure how motivated the insurance companies are to make make this big investment in technology for sensors, etc.  If the tech improves pricing efficiency then the consumer gains/ looses by more predictive accuracy of risk but it does not increase the profitability of the pool for the underwriter so there's not much incentive to spend the money.

 

In addition if the current models are pretty good at predicting risk (I think they are) and there is marginal improvement in pricing accuracy -say 5-10% increase in accuracy - the cost of tech won't be worth it for underwriters.

 

I think these things happen gradually over time, and it could be that the customers buy the sensors not the insurance companies.  Insurance company provides a potential for up to $S discount for buying and using sensor X, more and more people take up the offer.  Then the same for sensor Y and Z and so on.  You can see this in the driving sensors that you can have installed on your car.  Obviously good drivers will do this, but also some bad drivers may opt to do this and start driving more cautiously for an immediate effect on their rates that would otherwise have taken years of incident free driving to come down.  It can not only observe behavior, but change it.  Health sensors may be next. 

[eclecticvalue]

Guys check this out http://nypost.com/2016/09/21/peer-to-peer-insurer-lemonade-launches-in-new-york/ Daniel Ariely is involved with this.