This blog post has been five years in the making. Five years ago, we started exploring how our company could help with policy wordings. Immediately, we started hearing from commercial property underwriters. Can you help us review broker-manuscripted policies, these underwriters asked.
Commercial property insurance is unique in that many of the insurance policies are written specifically for customers by brokers. These broker-manuscripted policies put insurers in a difficult position, particularly in a soft market. What is included in each policy? This is a question underwriters must evaluate quickly in context of the risk and limits being adopted. For too long, review of these broker-manuscripted policies has been a chore because underwriters were not given the correct tools.
Based on thousands of hours talking to underwriters and research and developing technology solutions, we have discovered a better way to review broker manuscripted policies and we want to share it with the insurance world. This system relies on a specific combination of human expertise and machine learning (a powerful type of computer software). Humans still determine “is this clause good or bad?” but the machine learning program first intervenes to determine if a particular underwriting rule applies to an insurance clause.
As you will see, this new approach to underwriting complex commercial property risks has two primary benefits. The first benefit is obvious: speed and accuracy. The second benefit is not so obvious, but reveals itself over time: insurance carriers can eventually understand an entire portfolio of policies and the inherent risk instantly, not just at a monetary level, but also at a contract wordings level.
I will explain how, but first let’s start with the why
Why Commercial Property Policies are Difficult to Review
I remember the first time we encountered commercial property policies at RiskGenius. We had spent the previous two years preparing our machine learning index based on casualty insurance forms. When we kicked off our first Proof of Concept (back in 2016), users started uploading Commercial Property policies. And I started learning about “Perils Included” and “Perils Excluded” and “Property Covered” and “Property Not Covered.” For someone who was used to simple “Insuring Agreements” and “Exclusions”, my team and I were confused and bewildered.
But once we figured out the structure of a commercial insurance policy, we quickly learned there was another layer of confusion for the underwriters trying to tackle a review of these policies. At first, it seemed like the property underwriters needed a simple redlining tool. The commercial property underwriters explained that they did not know how to efficiently review broker-manuscripted policies. Based on feedback from these underwriters, we focused on three issues:
- Where were clauses deleted?
- Where were clauses inserted?
- Where were clauses modified?
When brokers submit a manuscripted policy, they don’t tell the underwriter what has been deleted, inserted or modified. Initially, it seemed like simple redlining was enough to solve these problems. If you go into the archives of RiskGenius ideation materials, you will see that the initial solution we were going to create was called ContractRedline (creative!). And we started building a redlining tool.
But redlining has been solved by basic software that’s available on the market. Adobe and Word have built-in redlining tools. If you simply need to redline then you should be good to go with a tool that is probably already installed on your computer.
However, there was a deeper layer to the problems faced by Commercial Property Underwriters.
Why Commercial Property Requires More than Simple Redlining
Simple redlining only gets you so far when you are reviewing a complex commercial insurance policy. An underwriter can see what has been changed, deleted or inserted. But the underwriter still has to read the darn policy and determine if the clauses that exist are good or bad.
Insurance organizations have attempted to help underwriters determine what is good or bad by creating policy review checklists. Imagine a Microsoft Word or Excel document that lists rules to follow when you review an insurance policy. An underwriter uses this checklist to review the policy.
If I was a commercial property underwriter (maybe I was in a former life?), I would probably put the checklist up on one screen and the manuscripted policy on a second screen. Then I would read each clause of the policy. When I came across a new clause, I would determine whether there was a rule pertaining to that clause. Then I would fill out the checklist accordingly. And, if changes to the policy are necessary, I would then either edit the document, or, more likely (because it is a PDF), add comments.
That's a time consuming and inefficient process.
Imagine you are reading through the Perils Excluded section of a commercial policy that includes the following exclusion: “War or other violent events.” The underwriter then looks to his checklist and says to himself “Okay, my checklist says I must have an exclusion for war. I just read a sentence in this policy that sure sounds like a war exclusion. I’m good to go.” Then the underwriter checks off “War Exclusion” and moves on with his life that, unfortunately, involves continuing this policy review (I kid, sort of).
Another example: Imagine you have a rule in your checklist that states “Mandatory clause: Cyber Exclusion”. The underwriter is supposed to read through the policy and find a cyber exclusion. If the cyber exclusion is found, then check goes the box. If the underwriter doesn’t find it, the underwriter raises an underwriting red flag, usually via email to some higher up. And so on and so forth until the policy is reviewed and the checklist is complete.
(As a quick aside, stop and think about what happens to that checklist after the review is completed. Nothing. The poor checklist sits quietly, sadly, in an electronic folder. It only sees the light of day if something bad happens, like a claim, and someone needs to be blamed. Imagine what you could do with that data, particularly if the checklist items were tied to actual clause language. More on that later.)