# Corentin Derbré

## Algorithmic Insurance

###### 2019-01-10 00:00:00 +0800

My knowledge of the insurance world is extremely limited. The following is just a raw idea, and with certainty, the hardest would be the execution of this plan, not its thinking.

An online insurance that you can subscribe to and you willingly provide your personal data for lower prices. On the investor side, there are buffers of insurance, the closer to risk, the higher the return.

It only works when enough client data points are collected, over a certain ammount of time. Let’s say 100k clients over 4 years. These data can be bought to begin to save time.

In theory, it will be less expensive than any other insurance because everything is automated (except field agents).

### Clients

The clients would install something on their phone tracking their location and their browsing or installed apps plus manually provide info, like “do you have a vehicle?”. The insurance can be turned on or off at will (with a cool-in time to avoid abuses?).

The client pays for what he will use the insurance for statistically. He insures a 100€ phone. There’s 1 chance out of 1000 to get it stolen per year. So he pays `100 * 1/1000 + fee = 0.10€ + fee` per year.

The fee is there to provide the service. Let’s assume it’s 0.05€ per year for the phone. 0.01€ would go to the company and the remaining 0.04€ to investors.

### Investors

The investors are an insurance of insurance. They agree to provide money when the company isn’t able to repay, when reality is above/under the statistics. There are many levels of investors, each being paid according to the risk they take.

Let’s say 2/1000 phones get stolen a specific year, and 1000 users are insured. The company is short of 100€, the cost of an extra phone. That’s where investors, who usually benefit form the fee, come into play to repay this phone. Because this sometimes happen (hopefully less than once every four years), they get paid 0.03€ of the fee.

Now let’s assume 5/1000 phones get stolen. The company’s data is off by a magnitude of 5. In this rare but possible event, second level investors agree to pay for all the phones. These second level investors get 0.01€ of the fee, because this is a rare event (supposed to happen only once every 50 years or so).

For this example, first level investors would pay up to 2x magnitude, and second level investors up to 5x. Above 5x, the company should close doors or seriously review thir data. A real scenario would involve much smaller magnitudes and more layers of investors, and different numbers than the ones that I provided.

A small thing that could yeild a lot of good press is if 1/10000 gets stole (less phones stolen than planned) on a specific year. The company could give back a little bit of the money to the clients, not even in proportion to the results. Word would spread. The investors still get their fees.

This way everybody is happy, clients pay as low as algorithmically possible, investors earning money at their risk level, and the company running whatever happens.