For every risk, there are countless variables to weigh against the price of insurance to determine if the rate will be adequate. Unfortunately, human underwriters can’t identify all factors and make a rate adequacy judgment on an individual policy basis in less than a second.  

Fortunately, machine learning makes it possible.

Driven by machine learning, a SaaS point-of-sale risk scoring platform can analyze a multitude of variables, and how they intersect, and compare them to the filed rate in mere milliseconds.

Now, for the first time ever, underwriters can turn on all segments and select risks one policy at a time … if they have the right platform. Let’s take a look at how this works using the case of the red Corvette.

The All or Nothing Quandary

To see how machine learning can improve auto insurance risk selection, consider the case of an underwriting team that has observed high losses with red Corvettes.

This team makes the decision to stop offering quotes for all red Corvettes based on two variables – vehicle color and make – without also considering the question of rate adequacy.

Now, Andy, the 28-year-old male Corvette driver, with two accidents and a speeding ticket will never receive a quote. But what about Flora, the 67-year-old grandma who uses her red Corvette for Sunday drives and has never, ever had a claim? Unfortunately, she won’t receive a quote either.

This is a case of risk selection gone wrong.

• First, how does the insurer know if their rate for these risks is inadequate? Andy could be a great risk if the insurer’s rate is high enough.
• If the rate is inadequate for Andy, why is Flora excluded based on Andy’s mistakes?
• Assuming the rate is adequate for Flora, why is the insurer still forfeiting Flora’s good business?

Here’s the reality: Human underwriters are limited to a finite number of factors when deciding which risks to accept and which to decline. In this case, the red Corvette underwriting rule is applied universally to all red Corvette drivers, even though other more important risk characteristics may have been present in some. These decisions were also made without consideration given to rate adequacy.  

Even though red Corvettes may be correlated with high losses overall, it’s unlikely that all drivers of red Corvettes are inadequately rated. The color and make of the car should be weighted differently for drivers with different risk profiles and the combination of different risk characteristics present within each profile.

Unfortunately, in this scenario, human underwriting teams typically don’t have the time or the tools to dig deep enough to accurately separate adequately-rated and inadequately-rated risks on an individual basis, so this all-or-nothing strategy seems to make sense – or at least it did before machine learning became a viable alternative.

How Machine Learning Shifts Risk Selection Into High Gear

In the same scenario, machine learning can assess a multitude of variables in milliseconds. Furthermore, machine learning understands how risk characteristics intersect to determine which red Corvette drivers are prone to losses. And, all these factors are compared to the rate to provide a rate adequacy score for each policy.

Using our previous examples, the machine learning algorithm may determine that …

With this policy-specific scoring, insurers don’t have to turn off the entire segment of red Corvettes. Instead, they can weed out the inadequately-rated risks while keeping those that have rate adequacy.

In some cases, the retained risks could be as high as 80% or more of the previously turned off segment! This innovative risk selection method enables the insurer to capture more earned premium while driving down loss ratios.

Humans excel in many ways. They have more creativity, empathy and common sense than any machine. However, when it comes to analyzing large sets of data, machine learning wins every time. The advantages are so extreme that not using machine learning in underwriting has become a major competitive disadvantage.

But It’s Not Just Machine Learning – Having the Right SaaS Matters

While machine learning is game changing, it’s only effective when it’s correctly leveraged and put into the context of underwriting. Furthermore, you need an algorithm to analyze your own dataset and a platform that enables you to set your own thresholds to deploy your company’s unique underwriting formula.

Fortunately, Soteris has undertaken the hard work of creating a proven and ready-to-use SaaS platform that achieves all these goals and more.

Now, carriers and MGAs of all sizes have access to a solution that is the result of years of data science, engineering, testing and refinement.