In the world of InsurTech, there are distribution players and there are data players. The data players are essentially doing two things:
First, they are enabling and exploiting new sources of data, such as telematics, wearables and social listening.
Second, they are processing data in completely new ways by applying data science, machine learning, artificial intelligence and high-performance computing.
The result is that, for insurers, the InsurTechs are creating opportunities for the development of new products for new customers; improved underwriting and risk management; and radically enhanced customer engagement through the claims process.
Which is why, in my humble opinion, tech-driven innovation in insurance will be data-driven.
As a result, this week I feature an Israeli start-up called Atidot, a cloud-based predictive analytics platform for actuarial and risk management…aka, the next gen of data modeling and risk assessment!
I’ve recently Skyped with CEO Dror Katzav and his co-founder Barak Bercovitz. Both have a background in the Israeli military, where they were in the technological init of the intelligence corps. Both have a background in cyber security, data science and software development.
These are two very smart cookies!
And they have applied their minds to the world of insurance and, very specifically, to data. To change the way that data is cut and diced to provide multiple insights from very different perspectives has been their purpose.
The result is Atidot, which in Hebrew means, “fortune telling.”
What’s the problem?
Dror explained it to me:
“Insurers (or rather, actuaries) are not doing all that they could with the data they have. And there are several reasons for this.
“First, they miss the point, Insurers look at data from a statistical perspective and miss out on the insights and perspectives that can be seen from different points of view.
“Next..., the traditional modeling tools that are still being used today are cumbersome, difficult to re-model and rely heavily on manual effort. With new sources of data now available, these tools are simply inadequate to handle them.
“And third, they’re too slow. The frequency of updating the models is too long, measured in weeks and months. This is because many of the current tools are limited in scale and flexibility, unable to cater for the huge volumes of data now available to them.”
How is work done today?
Today, insurers think about key questions to ask prospective policyholders. Do you smoke? Do you drink? Do you have diabetes? What is your gender? What is your location?
Insurers map the customer’s answers onto a statistical table. This linear modeling approach provides a risk rating of a certain outcome, such as the mortality rate for a life product.
But data science does not follow a linear model. It is different and varied. Data is modeled to show different correlations of risk to key variables.
This is what Atidot does.
It applies multiple approaches simultaneously to process a much larger set of data. This will include existing data that was previously ignored, such as the day of the month the salary is paid or frequency of ATM withdrawals, through to new sources of data, such as driving behavior or activity levels.
And while it is still very new for insurers to link, for example, increased levels of activity to mortality rates, there is enough evidence to suggest that it is just a matter of time before they do. You only have to look at the number of competitions on Kaggle to see that!
This shift gets to the crux of the insurer’s problem:
Quite simply, traditional models don’t have the ability to handle the new sources of data. Nor do they have the muscle to process it.
I’ve previously covered some brilliant InsurTechs in the data space, including Quantemplate and Analyze Re. FitSense is a data aggregation platform that provides insurers with a new source of data to underwrite life risk differently. The platform collects data from all major fitness and activity tracking devices. The data is then normalized (to weed out differences in the way activity is tracked) and presents the underwriter with a common score to indicate activity patterns and levels (just as Wunelli enables a driver behavior score from telematics data).
However, the challenge for insurers is knowing what to do with this data and how to handle it.
Dror put this into context for me:
“Let me give you an example from a South African life company who were building two life products – accidental disability and severe infection disease. To test our platform, we ran their traditional method alongside ours.
“We found that they had a lot of data about their customers that they were not using or taking advantage of. And even if they tried to, the actuaries did not have the means to group this data and properly assess it in their models.
“Atidot were able to group the data differently using our tech and show them how they could significantly improve the accuracy of their forecast tables.
“We showed them how they could look at data in a different way.“
This all sounded great, so I pressed Dror for examples and we started to talk about a piece of data that seemed irrelevant to a life risk assessment – the day the premium is collected.
Dror showed me a sample of data from a live pilot the company ran for a U.S. life business on a 50,000-customer sample.
It showed that customers who paid their premiums on the 14th of the month had a 20% lower lifetime value than those who paid on the 1st.
By enabling multiple data models to run simultaneously and picking the best model to better understand customers, Atidot drew a relationship between data that the actuary didn’t have before. Nor would the actuary have intuitively thought of it or arrived at it through a linear modeling approach.
So, is this enough to change the way insurers rate risk? Or change the risk selection criteria for an insurer?
To answer this I turned to Alberto Chierici, co-founder of Safer and an actuarial consultant with Deloitte. He told me:
“One issue to overcome for insurers is communication to the customer and regulators. For example, in some states it is compulsory to communicate to consumers why and how rating factors (gender, age, ZIP code) are used in pricing.
“That is making many insurers reluctant to adopt machine-learning-based risk rating and pricing. Think about the example you cited about people paying the 1st of the month versus people paying the 14th – how do you explain that to customers?”
Alberto pointed me to this discussion on Kaggle to illustrate the point.
One thing is clear, the InsurTech puck is heading Atidot's way.
The original version of this article appeared here.
The result is Atidot, which in Hebrew means, “fortune telling.”
What’s the problem?
Dror explained it to me:
“Insurers (or rather, actuaries) are not doing all that they could with the data they have. And there are several reasons for this.
“First, they miss the point, Insurers look at data from a statistical perspective and miss out on the insights and perspectives that can be seen from different points of view.
“Next..., the traditional modeling tools that are still being used today are cumbersome, difficult to re-model and rely heavily on manual effort. With new sources of data now available, these tools are simply inadequate to handle them.
“And third, they’re too slow. The frequency of updating the models is too long, measured in weeks and months. This is because many of the current tools are limited in scale and flexibility, unable to cater for the huge volumes of data now available to them.”
How is work done today?
Today, insurers think about key questions to ask prospective policyholders. Do you smoke? Do you drink? Do you have diabetes? What is your gender? What is your location?
Insurers map the customer’s answers onto a statistical table. This linear modeling approach provides a risk rating of a certain outcome, such as the mortality rate for a life product.
But data science does not follow a linear model. It is different and varied. Data is modeled to show different correlations of risk to key variables.
This is what Atidot does.
It applies multiple approaches simultaneously to process a much larger set of data. This will include existing data that was previously ignored, such as the day of the month the salary is paid or frequency of ATM withdrawals, through to new sources of data, such as driving behavior or activity levels.
And while it is still very new for insurers to link, for example, increased levels of activity to mortality rates, there is enough evidence to suggest that it is just a matter of time before they do. You only have to look at the number of competitions on Kaggle to see that!
This shift gets to the crux of the insurer’s problem:
Quite simply, traditional models don’t have the ability to handle the new sources of data. Nor do they have the muscle to process it.
I’ve previously covered some brilliant InsurTechs in the data space, including Quantemplate and Analyze Re. FitSense is a data aggregation platform that provides insurers with a new source of data to underwrite life risk differently. The platform collects data from all major fitness and activity tracking devices. The data is then normalized (to weed out differences in the way activity is tracked) and presents the underwriter with a common score to indicate activity patterns and levels (just as Wunelli enables a driver behavior score from telematics data).
However, the challenge for insurers is knowing what to do with this data and how to handle it.
Dror put this into context for me:
“Let me give you an example from a South African life company who were building two life products – accidental disability and severe infection disease. To test our platform, we ran their traditional method alongside ours.
“We found that they had a lot of data about their customers that they were not using or taking advantage of. And even if they tried to, the actuaries did not have the means to group this data and properly assess it in their models.
“Atidot were able to group the data differently using our tech and show them how they could significantly improve the accuracy of their forecast tables.
“We showed them how they could look at data in a different way.“
This all sounded great, so I pressed Dror for examples and we started to talk about a piece of data that seemed irrelevant to a life risk assessment – the day the premium is collected.
Dror showed me a sample of data from a live pilot the company ran for a U.S. life business on a 50,000-customer sample.
It showed that customers who paid their premiums on the 14th of the month had a 20% lower lifetime value than those who paid on the 1st.
By enabling multiple data models to run simultaneously and picking the best model to better understand customers, Atidot drew a relationship between data that the actuary didn’t have before. Nor would the actuary have intuitively thought of it or arrived at it through a linear modeling approach.
So, is this enough to change the way insurers rate risk? Or change the risk selection criteria for an insurer?
To answer this I turned to Alberto Chierici, co-founder of Safer and an actuarial consultant with Deloitte. He told me:
“One issue to overcome for insurers is communication to the customer and regulators. For example, in some states it is compulsory to communicate to consumers why and how rating factors (gender, age, ZIP code) are used in pricing.
“That is making many insurers reluctant to adopt machine-learning-based risk rating and pricing. Think about the example you cited about people paying the 1st of the month versus people paying the 14th – how do you explain that to customers?”
Alberto pointed me to this discussion on Kaggle to illustrate the point.
One thing is clear, the InsurTech puck is heading Atidot's way.
The original version of this article appeared here.
