Words That Bind
Insurance policies have no shortage of clauses and endorsements. Insurers, however, often have little idea of the promises being made or the exposures they are covering.
- Kate Smith
- May 2019
CLOSER LOOK: Chris Cheatham (left) of RiskGenius and Ted Stuckey of QBE Ventures discuss how technology has enabled QBE to examine policy language across its portfolio.
When QBE Ventures launched in 2017, managing director Ted Stuckey was looking to invest in startups that could make a near-term impact on QBE.
His first partnership was with RiskGenius, a Kansas City-based technology company that uses machine learning to compare wording used in insurance policies.
RiskGenius' platform struck at the very core of insurance—policy language.
“I think it makes a lot of sense that RiskGenius was our first partnership, because it solves really a fundamental problem of the insurance industry: Understanding the promises that we make to our insureds through the words we have in our policy documents,” Stuckey said. “What RiskGenius allows us to do is get a better understanding of: What are those promises? What are those words? What are the clauses that we're putting in there, and how can we better analyze them, get insights into them, and then make better decisions so that ultimately we can create better products?”
RiskGenius turns documents into data, then analyzes that data to compare and catalog the clauses and definitions used.
It didn't start out as an underwriting tool, however; it started as a claims solution created by two frustrated attorneys. Chris Cheatham was one of them.
Cheatham developed a disdain for messy insurance documents while defending surety companies against bond claims. He and fellow attorney Doug Reiser set out to build a platform that would collect, organize and deliver claim documents to attorneys, consultants, accountants and claim handlers. They then pivoted into policy analysis.
In 2018, RiskGenius expanded into a new area—regulatory compliance. RiskGenius developed software to streamline the form-filing process, with the goal of expediting the turnaround time needed for examiners to review products and ensure they're compliant with all state regulations. It is working with 10 departments of insurance to pilot the web-based platform.
“We expanded into the regulatory compliance market because regulators asked us to,” Cheatham said. “It started with the Iowa insurance commissioner suggesting to me that there was an application for RiskGenius within the System for Electronic Rates &Forms Filing [SERFF] system.
“After a year of discovery, we realized that RiskGenius can help with the filing process in a number of ways. First, carriers and states can review regulations and discussions next to relevant clauses in our software, something they could never do before. Second, carriers and states can research similar clauses that have been accepted or rejected in the past. We are excited to bring these innovations to the filing process and the 10 states we are piloting with.”
AMBest TV spoke with Cheatham and Stuckey to discuss how RiskGenius converts insurance documents into data and tools for the industry.
Below is an edited transcript of the interview.
Chris, can you give us a glimpse of how RiskGenius works?
Cheatham: Just think about a document that has lots of clauses in it, and if it's an insurance policy, probably there are a lot of them. You have thousands of endorsements, because of how you have to attach policies. What our software does is, it takes that document with all those clauses. It identifies the clauses, and then it labels them—“exclusion-war” or “definition-asbestos,” or lots of different variations. Then it stores this big clause index, where then the user can go in, grab a document, or multiple documents, click a button for definition of war and see it right away in that policy. Or they can compare their definition of war to other versions of definitions of war in their database. It's a lot of understanding of language at a very elemental level that hasn't really been done before.
Is there a lot of inconsistency in language within these policies?
There was one example where one carrier we talked to had 100,000 forms. They didn't even know how many variations of the definition of environment they even had. A lot of times, you don't even know what you have.
For example, with the opioid crisis right now, what language do you have in your policy that could affect that? There's no great way to figure that out right now. If you're just storing your documents as individual documents, you have to go through each one, one by one. Whereas you can pinpoint a clause very quickly in our system and say, “All right, which of the policies have that exclusion for pharmaceuticals and which ones don't?” Then you can start picking up what you have in your policies, and understanding your risk.
Why is it important to be able to compare the policy language? What's at risk?
Cheatham: Number one, you may want to be creating new products. If you're in the property group, and you need to start including a cyber exclusion, you may ask: “What other cyber exclusions have been written within our organization before?” You can go in and find those very quickly.
Another example is looking backward (at bound policies) and figuring out, for example: What have we included in them for cyber, or are we silent on cyber? We can figure out if a judge rules in the future that we don't have a cyber exclusion and so we're responsible, how many of those policies are out there.
Stuckey: It's a quintessential enterprise risk management conversation. We're writing these policies.
We're making these promises to insureds. If we don't have a firm grasp on, “What are those exclusions? What are those definitions? How loose or restrictive are these policies?” then we could be on the hook for a pretty massive loss if we're not careful. As an organization, it's our responsibility to understand, both for our insureds, for the regulators, for the shareholders, what we're doing. What are we writing?
A product like RiskGenius allows us to get those analytics. For us, it's taking it a step above what Chris talked about, where you actually get the clauses themselves, and it's looking at the analytics. It's saying, “What percentage of the policies that we write have this master clause in them? What percentage of them allow for these types of losses, versus don't allow?” Then we can take a step back, look at ourselves, and look at the way that we're accumulating the risk, the way that we're writing risks, so that we're better positioned when we ultimately have to make a claim payment.
Is there some kind of a competitive edge or a competitive risk to not knowing that?
Stuckey: I think definitely, as an industry, we're better off if we're not hiding behind our language. We're better off if we're much more open about the policies that we write, and what's in them. From our perspective, certainly, it allows us to be more responsive in times of massive disasters or massive needs. If you think back to Sept. 11 and right after that, all the carriers started going through to understand which of their policies had war and terrorism exclusions, and what was the actual language of those, so that they understood what their exposure was to that.
If you actually take a step back and try to understand, “What is that effort that's required to do that?” it's literally a bunch of lawyers looking through page after page, word after word, and doing that analysis by hand. That's not sustainable as we continue to innovate with new products, as we continue to scale as carriers globally and from a product perspective. You need a platform like this to do that, otherwise it's not possible for some carriers. It's not really an option to not do it. You really do need to understand those promises that you're making.
Did you run some kind of a proof of concept with RiskGenius?
Stuckey: Absolutely. The minute we introduced RiskGenius to our business partners within our North American organization, there was an immediate connection that the problem they were solving was a very real problem.
Then certainly, the next question is, “How good is it? How quickly can we turn this around?” We ran, I would say, a relatively intensive proof of concept with RiskGenius—something like 8,000 documents, and probably close to a dozen users across the U.S. And we literally went in and manually understood, “What's the time difference between this task and that task with RiskGenius? How much more quickly can we answer those questions from the field? How much more accurately can we understand our exposure, or our units at risk?” Ultimately, it got to the point where we recognized this is the product that we want to move forward with and deploy more broadly across all of our business units in North America.
Can this be used in underwriting as well?
Stuckey: Yes, we see use cases across the board. Obviously, there's a massive lift for the product development team, but from an underwriting perspective, it arms you with so much more information at the point of sale. Whether you're talking to a broker about why your product is superior, you're giving more insights to a broker or an insurer around what's covered and what isn't. On the claims side, it's very efficient for them, then, to look back into that policy and understand what's excluded and what isn't at the point of first notice of loss. We see use cases all across the organization. Then also at a groupwide level, where you're doing more of those analytics around, “What's our cyber exposure? What's our war and terrorism exposure? What's our asbestos exposure?” Things like that, that we really can't get a fine level of detail or insight on without a massive amount of effort today.
Is it machine learning that is powering this?
Cheatham: Yes, the machine learning's really looking at those clauses and saying, “All this content in this particular clause, we think it means the same thing as this other clause that we've already looked at before. This other clause we've already looked at before, we called it a war exclusion, so we're going to call this one a war exclusion, too.” It's using words, sentences, and the parameters around all those words to score them against everything else in the database.
Stuckey: It's crazy to think that you need machine learning to do this, until you actually understand what it's doing. If you look at an insurance policy, actually capturing what specific clauses are is not straightforward. You need some intelligence to break those clauses out, break those definitions out, start to draw a correlation between multiple clauses and things. That's where a lot of people don't fully appreciate the power behind this and the problem that it's trying to solve.
It's not just taking all the words and dumping them into a Word document. It's actually structuring out individual clauses, allowing for redlining to happen in between clauses where there's gaps and stuff. That's some pretty smart stuff.