They have existed since the beginning of the 19th century and their business has continued: insurers have spent years evaluating the potential risks of things and people, offering a service that they will try never to give you while charging for it.
They were the first to deal with Big Data and will also be the first to integrate into their risk-benefit analysis the prediction of traffic accidents through Google Street View, a model developed by two researchers from Stanford University and Warsaw.
The study was registered on arXiv (a repository of scientific articles) at the end of April. In it they defend how, using the photos that Google Street View provides of every street and almost every building in the world, they have been able to bypass the standardized model that most insurers use to calculate the probabilities that a driver has of suffering an accident. accident.
Street View photos may be indicators of something. At least that was demonstrated in 2017 by another team of researchers, who managed to classify 22 million vehicles based on images taken by Google’s satellite. Then they established a relationship between the type of vehicle that each person has and their level of income, education or occupation, composing a sociodemographic map of the US in a matter of a few days.
Now, instead of extrapolating the photos of the cars towards something bigger, the researchers have taken as a reference the homes of almost 20,000 Poles who between 2013 and 2015 took out an insurance policy for their car. The study takes place in Poland, a country with just over 38 million inhabitants and in which insurers abound as much as allergies in spring.
Your home, indicator to know how you drive
The researchers say that having the photos of each home, a series of variables can be extracted that help insurers to calculate the risk of the policy. These characteristics are the type of house where the potential insured lives (a villa, a semi-detached house, a block of flats), the age of the house, its conditions, an estimate of the purchasing power of its members and what the rest of the household is like. neighborhood (in terms of the density of homes it has and its type).
“Unlike most products, the ultimate cost of an insurance policy is not known until the time of sale. It is, therefore, a challenge to establish an adequate price and insurers try to take advantage of statistical methods to predict the future risk of each client in advance”, continue the authors of the study.
Anyone who wants to take out car insurance right now will have to answer at least five questions: how old are you, how many years have you had your license, have you had accidents in the past, where do you live and what is your car. With her answers, the insurer will make various statistical projections and try to predict whether or not you will be profitable in the future.
The geographical factor is important. It is not the same to drive in Madrid as it is in Albacete, nor in Warsaw as it is in New Delhi, which is why companies tend to increase the price of insurance depending on the location. That model is fine except that it assumes that everyone drives the same way and has the same probability (just because they live in that place) of having an accident. And this is where house photos come into play, adding another layer of information for insurers to make their predictions.
The problem of privacy and third parties
“We found that visible features in a photo of a house can be predictive of car accident risk, independent of classical variables such as age or zip code,” the researchers say. By incorporating the new level of information into current risk calculation models, they found that it improves their predictions by 2%.
To carry out the study, a Polish insurance company shared 20,000 customer records with the researchers, which contained the personal information of each one, including postal address. From there they were able to pull the photos from Google Street View. However, the model raises several questions about privacy.
By signing an insurance policy, a customer agrees to share their data with the insurer. What is more doubtful is that he will also accept that this one can search for photos of his house through Google. In some countries, household characteristics can be indicators of a person’s proximity to a cultural group, ethnicity or religion; For this reason, when insurers take this information to reduce or increase the price of one of their products, they would be discriminating based on these conditions.
The study is still in the process of scientific review, but the researchers do not rule out that, after insurance companies, the next sector to adopt this model is banking.