Exploring Use of GenAI in P&C

 

Tom Kerr (TK): Mainstream use of generative AI has prompted claims leaders to explore how the technology can be used to improve claims management. And while GenAI certainly offers tremendous opportunities for the P&C industry, tech leaders must be diligent in ensuring the intricate knowledge and experience in claims management is integrated into these programs.

In today’s podcast, we invited Enlyte tech experts Mike Bishop and Mike Cwynar to discuss these matters. Gentleman, welcome to show.

Mike Bishop, let’s start with you. When you hear from customers regarding incorporating GenAI into their claims programs, what are they typically looking for?

Mike Bishop: I think right now what they’re doing is what everyone is doing, which is trying to understand the technology. So, it’s been hyped, it’s come on the scene very recently, and I think everyone sees a great deal of promise in the technology, but they’re trying to understand how to use it.

And that’s the hard part with any of these technologies, right? It’s one thing to look at the technology for technology’s sake, but to actually incorporate it into a workflow, especially a workflow that is involved in treating injured employees, people injured in automobile accidents, you have to be very careful about how you use it.

So, I think right now what customers are doing is trying to understand the kinds of use cases, where it would be applied and also, to understand how they can use the technology responsibly.

TK: Do they come with a general idea of how they’d like to incorporate the technology in their claims management programs, or is there a learning gap that still exists?

Bishop: I think there’s definitely a learning gap. Certainly, they can come with ideas, the latest press release that they’ve read, or they can come up with ideas of technical things that you could do with it, right?

Like, "Oh, I could summarize this, or I could have a chatbot that does that," but nothing specific. And I think, again, they’re just trying to understand better what types of things are truly possible, and then again, how they’re going to incorporate them. So, I think there absolutely is a learning gap.

Mike Cwynar: Yeah, I think most come in with a list of problems that they think that they can solve with this just given an understanding of their businesses the potential behind GenAI.

As businesspeople, I don’t know that they always know where to start. They just have a problem. And to Mike’s point, some recent announcement sounds like maybe it would check that off their list. Can I predict quickly whether a claim’s going to need something longer term or not? Today, sometimes they don’t know until later on in that cycle. Tomorrow, they’d like to figure it out sooner rather than later through the use of historical data.

Many just don’t necessarily know where to start and I think that’s where the learning gap comes into play. Because it isn’t necessarily as easy as it sounds on paper to figure something like that out.

TK: And to continue on that theme, Mike Cwynar, what other challenges do claims industry professionals face in better understanding and implementing this technology?

Cwynar: I think with data, quite frankly, folks are getting a little more educated about it. You start asking questions like, “do you have the data for this? When’s the last time you cleaned any of it up?”

What you end up hearing from a lot of, at least the larger payers, is data in multiple systems. It’s not necessarily that easy to get to. And I think just having a lot of data doesn’t necessarily even solve the problem. The right amount of data is going to be important for building and training some of these models.

And I think the last part of it really is who within the organization is the subject matter expert that can help validate that these models are producing answers to the questions the same way that their business typically would? That’s the power behind it. Every payer can have their own sort of unique way of settling claims and handling underwriting, etc. And, so, the power of GenAI is designed to let them have their own philosophy built into these decisions.

But it takes time to train these models, and these models need data. And, not all data is good data. And so, I think that’s where a lot of folks are really starting to put more focus in here, realizing there’s a lot of entities out there coming with these potentially powerful models, but they’re not any good without customer data.

Bishop: I think the other one, too, that is not as well appreciated sometimes is the fact that the technology providers that typically serve our industry and serve all industries, are also trying to learn these technologies and they’re really not ready.

Whenever you meet with the technology providers, they’re still trying to figure out how they’re going to make some of these core technologies available to us. So, it really hasn’t settled down where you can sort of make a technical choice.

Most people cannot rationalize going out, for example, and training a large language model on their own. They’re going to build off of one, specialize it through various techniques, including prompt engineering.

But because the technology providers really aren’t ready right now, you’re almost in a wait‑and‑see mode where you’re trying to figure out who’s going to come to market with the best tools. And so, until that plays out, I think that’s another obstacle to bringing this stuff to market.

TK: And, I think that's a good transition to our next topic. What questions should payers ask when they're selecting a GenAI tech partner?

Bishop: First of all, you want to make sure that the tech provider appreciates the particular challenges in our industry. One of the euphemisms that they have around these large language models is the errors that can come up. They call them hallucinations or various terms, which really just means the model is spitting out something that’s just wrong, that’s made up. And so, you have to deal with all those things.

If that happens when you’re trying to figure out something that you’re buying online or something where the ramifications are small, it’s not as big of a deal. If you’re providing health care information to someone, the ramifications of that could literally be life‑threatening. And so, you have to make sure that the people that are providing the models, that are providing the technology, are taking into consideration how you’re going to responsibly use the technology. And to make sure that the guardrails are there.

Things like biases can creep into these models. Just making sure that they’re not just using technology and sort of viewing it that way. That they are thinking about how to apply it into our industry.

TK: What are some strategies payers can follow to ensure they're getting the best results from GenAI?

Cwynar: I think one of the things that we’ve been talking to many about is having a very clear understanding of the problem statement and how you measure that, you’re getting the outcome that you want.

So, like Mike was just talking about, the ability to monitor for answers that don’t make sense. For example, there’s a model that’s running that’s helping an adjuster make decisions that have some regulatory compliance background in them, and all of a sudden, a new fee schedule in Florida comes out or Michigan comes out with some of these models that could potentially become irrelevant overnight.

So, I think one of the biggest things is to really make sure that, in all cases where you’re running this, that some amount of subject matter expertise, human in the loop, is involved to, if nothing else, periodically audit and look at the results that are coming out of the models to make sure that they continue to be relevant.

Because these aren’t one‑and‑done things. We don’t build it once, and then they just kind of work in perpetuity. You have to be regularly on top of this stuff, which then gets into good data hygiene, right?

Same thing with data. It’s going to change over time, and so you got to just be really on top of that kind of governance model to make sure that these models don’t start providing answers that potentially become irrelevant at some point in the future and no one really understands it.

TK: So, in terms of score-carding how well GenAI works in the claims management space right now, are payers mostly looking at how well it can make processes easier or make claims handling more efficient? Or are there are other goals they want to achieve?

Cwynar Yeah, it’s like progress over perfection. You don’t necessarily need to tackle the most complicated problem right out of the gate. Start small, get comfortable with it, understand how to monitor, control and audit it.

Because the possibilities of what you can apply this to become somewhat unlimited within the world of claims, when you think about interacting with policyholders, injured employees, etc., to potentially how it influences underwriting to fraud detection, it’s in there. But, it’s reliant on people that really understand the business and the ability to keep an eye on what’s going on day to day.

TK: Do you have any predictions on how GenAI will impact the industry in the next one to five years?

Bishop: I think you’ll see it in particular areas. And I think, this has been a thread that has been in almost all of the answers that Mike and I have given, is that you have to think about the governance and how you responsibly use the technology.

And so for that reason, I think the impact that you’re going to see is less the stuff that is hyped and demoed; where you see an individual customer or, in our case, an injured worker directly interacting with the technology. I think the technology will have a bigger impact behind the scenes because then you can control it.

So I don’t see it happening in the next one or two years where we would provide a medical summary to a claimant, but we could certainly provide a medical summary that came out of GenAI to someone internal to make sure that they would look out for these errors, hallucinations, and those kinds of things.

So, it’s the responsible use requirement that I think, especially in our industry, is going to mean that the technology will be used behind the scenes for automation to sort of help knowledge workers do their job more efficiently and better.

Cwynar: Yeah, I agree. I think speed and efficiency become the biggest opportunities here. People are still going to file claims and they’re going to need help from time to time and you’re going to always want a human being available to have a conversation.

So, I don’t see this world where all of a sudden there’s a virtual adjuster who handles everything for you. That doesn’t mean that, to Mike’s point, behind the scenes, there isn’t a virtual adjuster that’s able to quickly and efficiently gather all the information that’s needed, provide recommendations, help sort further claims along much faster than they can today given the typical amount of work that sits in front of an adjuster.

I think that’s where you’ll see more of the assistant adjuster, the virtual adjuster component being pretty powerful for a lot of our industry today. And, I think the ability to be able to quickly triage, and access information that otherwise wouldn’t always be available. Say if somebody just filed a claim and they’ve had prior accidents, which would alter the course of the care that somebody needs to be able to identify things like that upfront.

It just creates more of a personalized plan of care very quickly and early on, which then gets people better and off to their lives faster. Those are probably some of the areas that many are talking about and focused on, but I think have the biggest potential over the next couple years.

TK: Thanks, Mike and Mike. And we’ll be back with another podcast soon. Until then, thanks for listening.