RiskGenius Blog

Insights into our world of Insurance & Technology.

Size *Sometimes* Matters: Using cyber policy length and similarity score to reduce drafting risk

The following series of posts survey how the seemingly nascent features of an insurance policy - policy metadata - can be levereaged to create artificial intelligence that improve the quality and efficiency of drafting these policies. By using even the most rudementary AI tools, underwriters and legal professionals can begin to speed up the work they are doing by more effectively spotting areas that deviate from industry standards.

On the face of it, something as limited as the number of clauses in an insurance policy would not be a helpful feature. However, this is one example of a feature that can be used in connection with other features to provide more transparency about the areas of an insurance policy that have historically been opaque and difficult to understand. For example, information about the ways usage of a clause changes across different lines of business can be mapped out; the way that different carriers refer to the same type of information can be more easily compared; and, perhaps most importantly, individual underwriters can gain a real-time understanding of whether or not they are drafting a policy that fits within a given regulatory framework or  their organization’s goals.

An Exercise

To show you what I mean, I am going to walk through an experiment where I look at two features from our cyber liability index and explain how AI can use even the most trivial of details as an oversight mechanism.

To start this whole process, our review team looked over approximately fifty cyber liability policies, broke those policies down into clauses, tagged them with information about the name of the clause, the type of clause involved, the carrier, and our data science team used a set of machine learning algorithms to score these different pieces against one another. This is how we are able to come up with attributes like average similarity score, which helps determine how common a specific clauses is based on comparisons to other clauses of the same type.


Looking at the average length of a cyber liability policy, we can see that there are only a handful of policies that have fewer than 50 clauses in them. Alternatively, there are also only two policies that have more than 150 clauses. 

Looking at the average similarity score from a cyber liability policy, we can also develop another set of information that can be helpful to underwriters.


From this it is clear that most policies are made up of clauses that have a similarity score of between 80 and 95, with only a handful of scores falling outside of that range. 
With this information alone, it would be difficult to say why size can matter. However, when we put those two sets of information together we can then see that some of the shorter policies have lower scores. Perhaps that is because they are not fully integrated. There could be other endorsements attached to the policies that boost the scores and which we were not able to get access to.

There are many possible explanations for why this could be. However, what is most exciting about such trivial AI is that, based on this aggregated knowledge of what people in the cyber liability market is doing, we now have a better idea of some of the things that might be causing a policy to score lower.

Looking at these two features together, we can help gain a more spatial appreciation for which types of policies are more effective than others:


And by working with other features to see what they can tell us, we can start helping underwriters draft and review policies with a higher degree of accuracy and more quickly than they could do on their own. For example, the above graph shows us that a lower number of clauses - fewer than 50, could be a strong indication that a policy is lacking 
something. This information also shows that really large policies are not always the most homogenous to other policy language. In fact, it might be that a larger number of clauses in the policy waters down the average similarity score. 

The really exciting parts, however, will come farther down the road as more pieces of information are layered on top of the basic features of a policy in order to understand what your insurance policy contains.

Coupled with information about the composition of a policy and most popular clauses, we could then take any given policy, compare the length, composition, and types of clauses included to the rest of our index, and identify valuable pieces of information. This information could then be used to identify the most frequently used clauses that are missing from the given policy or specific clauses within the policy that deviate heavily from industry standards.

This is why Artificial Intelligence has so much promise. It is not because these are incredible leaps that people did not have the power to make, but these are leaps that computers can make very easily, which yield great insights.