Overview

Historically, the most difficult part of correctly determining the pricing on an Insurance Policy, is in determining the various Ratings for both the primary policy and any additional riders (other specific cases) attached to the policy. This is compounded by the fact that even the base policy, which is often determined by a class of related individuals who share similar characteristics, may contain 10’s of different “known” factors, as well as many “unknown” factors.

While it is not a perfect solution for quickly determining the Rating on either the base policy or any riders, the application of Machine Learning / Artificial Intelligence (AI) can more quickly predict patterns, and thus the risk level for a given Insurance Policy. In this article we will discuss areas where AI is ripe for usage within the Insurance Industry, as well as changes in the overall industry that are harder to accurately predict.

‍

Determining the Rating of an Auto Insurance Policy

When determining the Rating of a Base Auto Insurance Policy involves many different factors. Some examples include the following, along with questions about modern social and technical developments.

Factors:

Gender (1)

• Males tend to be higher risk takers, thus a higher rating.
• How do we adjust for Transgender individuals?

Age (1)

• Individuals who are younger, tend to have less experience, and thus a higher rating.
• Likewise older individuals have slower reaction times, thus a higher rating.

Marital Status (1)

• Traditionally this is more important for a man, as it tends to reduce their willingness to take risk, lowering their rating.
• Does this factor still apply to same sex marriages?

Distance Driven per Year (2)

• The more miles or Km’s driven in a given year, the higher the rating.

Zip Code / Postal Code (3)

• This rating looks at the number of claims and types of claims within a given area.  This takes into account:  theft, break ins, accidents, medical, etc.
• However, a Zip or Postal Code can include a large area with different characteristics.

Primary Parking Location (4)

• Garage, Carport, Parking Space, or unmarked Space.
• These could affect theft / break-ins, weather effects, parking collisions, etc.

‍

Environmental & Weather Effects

• The frequency and type of environmental and weather effects:
• Extreme Cold or Heat (parts failures, causing accidents)
• Flooding
• Fog
• Haboobs (sand storms)
• Hail Damage
• Ice on the Roads (skidding, salt)
• Ocean Breezes (salt)

Primary Purpose of Vehicle (2)

• Normal vehicle used for commuting to work, local travel, vacations.
• Business vehicle - which will have a higher rating

Primary Driver’s History (1)

• How many moving violations does the Driver have?
• Has the Driver been found guilty of a DUI or other felony involving any vehicle in the past?

Vehicle

• Manufacturer
• Model
• Style
• Year Produced
• Safety Features
• Anti-Theft Features
• Newer vehicles will in general have better safety features, lowering rating for medical coverage..
• But newer vehicles also have more computers, carbon fiber parts, etc. which increases the rating for accidents.
• With older vehicles, the opposite is also true.
• Style or Type of vehicle - is it a sports car, a vehicle with a high center of gravity (more prone to rollovers), meant for off road conditions, etc. each of which may impact the rating.

Parts Availability

• Are OEM parts still produced by the Manufacturer?
• Are only refurbished parts or 3rd party parts available?
• Do parts need to be custom made, like with a really old or unique vehicle?

Color of Vehicle (1)

• Interestingly enough the color of the vehicle can make a difference in terms of the risk of an accident, moving violation, etc.
• Red, Yellow and Black are higher risk.
• Brown, gray, etc. are often lower risk.
• However, this can be impacted by the primary location of the vehicle.  For example, having a white car in an area where there is a lot of snow, reduces its visibility and increases the risk.
  

As well as many more.

(1) Naturally, one of the biggest technical impacts during the next 10+ years is going to be the continual development and adoption of “Self-Driving” vehicles.  As we start achieving Self-Driving vehicles of level 3 and higher, how important will some of these factors play into determining a policy rating.  For example, if the vehicle primarily drives itself, then is it really important that the policy is for a Male, 23 years old, and Single?

At the same time, what happens when:

• The auto-pilot is disengaged?  Can the vehicle measure this and report it back to the Insurer?
• The vehicle is driving through difficult terrain, dirt roads, or construction areas, where the vehicle may not be able to detect the road correctly or know what to do with oncoming vehicles.  Many dirt roads are one-way, so someone has to pull off to the side and wait.
• As we rely on a self-driving vehicle more and more, our driving skills will normally atrophy.  Then what happens when a person has to take control of the vehicle?
  

(2) Several Insurance companies already have devices that can be attached to the vehicle to determine speed, quickness of starts / stops, distance, and other diagnostics of the vehicle.  Now take this a step further, can the vehicle report this information and even the specific location to the Insurer when connected to either a satellite or cell phone tower?  Delivery companies already have some of these capabilities, but the Insurer could use this information to determine if the vehicle is speeding, making high speed turns, etc. which could affect the risk profile for the policy.

Likewise, does a person have to select an option describing what their trip involves, and can this information also be transferred back to the Insurer?

(3) Today it is easy to determine the Zip or Postal Code where the vehicle is normally kept, because it is part of the address information.  But what if we could start getting more granular - vehicles within X miles or Km’s from the exact address location?

(4) Today this is difficult to capture with any degree of confidence.  However, what happens when the vehicle is able to quickly take a 360 degree view of the surroundings of the vehicle when it stops and it is at its primary “Home” location?  Then apply Artificial Intelligence to this.

Odd but Apparently Fairly Accurate “Fun” Factor

As reported by Reuters. In 2007, Lee Romanov and her team from InsuranceHotline.com did a study, just for fun, of 100,000 Drivers in North America looking at 6 years worth of the Driver’s record and compared this to their Astrological (Sun) Sign. And the findings actually shocked Lee and her team. It ended up that the Driver’s Astrological Sign is a larger factor than Zip / Postal Code or even the Age of the Driver.

Worst Drivers:  Aries, Libra and Aquarius

Best Drivers:  Leo, Gemini, Cancer, and Scorpio

‍

Applying Artificial Intelligence

One we start collecting all of this data from the Vehicle that is being insured as well as others, either from other Insurers or from other covered vehicles within our portfolio; then we can start detecting patterns. And once we know these patterns and the correlation coefficient between various patterns, we can potentially quickly adjust the Rating of the policy.

For example, let’s take the Color of the Vehicle*:

• A Red colored vehicle by itself may not have a strong correlation to the number of accidents or moving violations (speeding tickets).
• But when the Color Red is combined with the Type of Vehicle:
• Red Sports Car - very strong correlation, increases rating
• Red Sedan - no correlation, no effect
• Red Light Duty Truck - moderate correlation, slight increase in rating
• Red Semi-Truck Cab - no correlation, no effect
  

*These are examples for demonstration purposes only.

And you could easily add additional Factors to this analysis, like the Manufacturer, Model, or even Zip / Postal Code, etc. in order to determine what combination of Factors is the best at determining what the rating for a given Base Policy should be.

At the same time, one must be careful of having too many factors used in a given calculation, as you have to use some weighting factor based on the data. This is another area where AI can assist, because it can quickly determine which combination of factors have the greatest impact on a given Policy Rating. For example, a 30 year old male, wants to insure his brand new Mercedes SL Roadster, which is red, he lives in Scottsdale, Arizona 85054, and has three speeding tickets within the past 6 years, etc.

Using this information now AI can compare this to other individuals within the same Zip / Postal Code, age range, similar type of vehicle, etc. in order to determine which factors have the greatest impact on the overall Policy Rating.

1. Gender + Age combination - small impact
2. Make & Model of vehicle - very large impact, after all this is a $ 92,000 car (base)
3. Make & Model of vehicle + Zip Code - small additional impact, as Scottsdale is expensive for repairs and also has a lot of similar vehicles; which could make a huge difference if the Policy is not a No-Fault policy.
4. Make & Model of vehicle + Red color - small impact
5. Make & Model of vehicle + Red color + Driver history - moderate impact, so this combination could replace the one above (#4), since the Driver already has a habit of speeding.
6. Etc.
   

By doing this, AI could determine a unique Policy Rating for a given customer, after looking at all of the data and information. Naturally, this is not going to happen all at once and in fact will need to be constantly tweaked and modified as new information comes available.

‍

What About Other Forms of Insurance

The same concept of using AI to look at the various data points and determine a Policy Rating can be applied to most but not all other types of Insurance.

Life Insurance

The traditional factors, like age, gender, smoker, overweight, cardiac issues, diabetic, length of coverage, amount of coverage, etc. would certainly still be considered as primary factors when determining a Policy Rating.

But as technology continues to advance, what happens when:

• Insurance companies start capturing or requiring Genetic Markers for susceptibility for specific diseases, types of cancers, etc. that could impact the life span of the insured.
• What happens if the Genetic Market test is wrong?
• How will we handle individuals who undergo Genetic Treatments to either remove a particular negative marker or to add a positive genetic trait to the individual increasing their life span?
• How will Insurance companies treat cybernetic enhancements or active prosthetics, how will these affect an individual, their life span, and thus their Policy Rating.

Property Insurance

The same is true when an Insurance company is determining the Policy Rating for a house, piece of property, factory, commercial store, etc.

Here again technology, along with environmental changes will have potentially dramatic long term effects on the Policy Rating for a given property:

• New construction materials, which are more fireproof, earthquake resistant, or more windstorm resistant.
• New safety devices, processes, use of robotics on the factory floor, etc.
• New anti-theft devices and tracking devices for stolen goods.
• Ocean incursion due to rising sea levels, causing flooding, salt water damage, or a complete loss.
• Weather pattern changes that impact the use of the property or increase potential damage to the property.  For example, in recent years “Tornado Alley” has started to shift from the Mid-Western States in the US towards more of the South Eastern States.  Is this a long term trend?  The same is true for Hurricane patterns and strength.
• Earthquakes and sinkholes caused by man-made impacts, such as fracking and draining underground aquifers, causing the earth above to collapse.

Unique Insurance

There will always be some items that AI won’t be able to determine an accurate Policy Rating, because the item or potential damage to it is very unique, and there is not enough similar data to determine what the overall risk is of damage, loss, or injury to others.  However, we would expect that over time these unique items become fewer and fewer.

‍

Conclusion

Artificial Intelligence (AI) and Predictive Analysis, in conjunction with new Technology and Environmental Impacts, will have a very significant impact on how Insurance Companies determine the Policy Rating for a given insured item, property, or person.

This change won’t happen overnight, but will represent an evolutionary advancement in how Policy Ratings are done.  Some will replace standard actuary tables, as AI becomes better at accurately predicting the risks for insuring something. It will be interesting to see how the various Regulatory Bodies make adjustments to take into account all of these things - from technology advancements to more accurate Policy Ratings.

Naturally, Insurance Companies who stick with the traditional Policy Rating models, will over time either be outsold by more accurate policies that use AI analysis; or worse take on more risks than they should, thus increasing the risk to the overall business as an ongoing corporation.

And if you don’t think this can potentially happen even with AAA rated insurers, one only needs to look back to 2008, when the massive Global Insurance Company AIG nearly failed and had to be bailed out.  Otherwise, there would have been a cascading failure among many of the largest banks and financial institutions globally.

We hope that you have enjoyed this article on how AI might be applied to Insurers and Policy Ratings, today and in the near future. To learn more about our AI solutions click here.