How A Long-Lived Super-Regional Carrier is Implementing AI in Claims
Identifying the needs and pain points of customers and then developing solutions to better service them and other company partners—those are the most important promises that artificial intelligence and new technologies hold for those overseeing an ongoing transformation at Westfield.
Westfield is a super-regional property/casualty insurance carrier based in Ohio that’s been around since 1848.
Despite its long history, the company recently has embraced modernization, and it is undertaking a technology transformation, as it were, by updating processes and systems with the help of artificial intelligence.
Claims Journal spoke about these undertakings with Andrew Quinn, product manager for generative AI at Westfield, and Jason Bidinger, associate vice president of claims strategy and technology.
The conversation has been edited for brevity and clarity.
Claims Journal: What new technology are you integrating into your claims department/processes?
Bidinger: I would say in terms of what we’re implementing, just from a broader perspective, and then focusing in that our claims area is focused on, is technology to improve our policyholder and claims professional experience. Our team has been working hard over the last couple of years to really understand our policyholder, our employee, our agent—their needs and pain points and then develop solutions to consistently improve that experience. So, we’re focused on digital capabilities for both agents, customers and internal stakeholders. We’re kind of focused around automation opportunities, but I think the really fun thing is over the last year or so, we’ve been digging into this potential of generative AI in our claims process, not only for our internal team, but also for our customers.
Quinn: I am the product manager for our generative AI utilization here at Westfield and so I have been heavily focused on just utilizing generative AI and specifically large language models and how we can integrate those into our various different applications across our organization.
CJ: Why are you implementing this technology? What problem is it solving?
Bidinger: I think when we think of generative AI opportunities, it’s pretty easy to see the potential to gain efficiency in our process, provide our adjusters additional resources in real time, to help them really do what they’re paid to do and that’s pay what we owe on claims. The other thing I would say, for our team, is the retirement factor. We have a large percentage of our claims team that is or will become eligible for retirement in the very near future. So, some generative AI solutions have the potential to bridge a significant knowledge gap. It’s a challenge to replace the amount of knowledge that we’re potentially losing, so this will help us kind of expedite some of that learning when it comes to the usage of our generative AI tools.
Quinn: I think a lot less about this is about being problems to be solved. Rather we see a lot of opportunities to improve both our employee and our customer experiences that can look like a wide variety of things—that can look like helping our claims representatives to sort through volumes of information more effectively, that look like providing better support to our customers through faster response times as we’re handling and adjusting claims. And it can look like upskilling our employees into new or different opportunities across our organization by enabling them to more quickly learn information and be better prepared to do their job.
CJ: How is it working?
Bidinger: I think we’ve seen some great potential. Our initial focus has been around summarization capabilities for claims, which, based on what I’ve seen, is fairly similar to what’s out there in the industry. We explored some potential vendor partnerships, but really found a great opportunity internally with Andrew and his team with some of the skills that we have in house—the ability to build out some products in a very efficient way.
Quinn: We have developed a few different products utilizing generative AI, predominantly in summarization, and at every point they have exceeded our expectations, whether that’s expectations on the accuracy of the summarization, or our expectations regarding the quantity of time we think these tools are saving our employees. As we’ve been able to implement these tools, we are more and more excited about the opportunities for us to continue to improve our experiences for both our employees and our customers.
CJ: How long has the technology been up and running?
Bidinger: In our initial use case we talked about summarization. Specifically, we started the focus on medical records and demand package summarization for claims for our casualty area. We also have explored some things around construction defect. I think we started experimenting back in the summer and then sometime around October we began to conduct a proof-of-concept and then a 60-day pilot with a prototype, which Andrew’s team developed using ChatGPT. We’ve successfully completed the pilot and now it’s really more about there’s some things internally. And in the claims world, obviously we have to be considerate of the regulatory compliance environment, so we’ve kind of built some internal checkpoints with teams within Westfield to make sure that we’re doing things the appropriate way, considering the appropriate things when it comes to the regulatory environment.
Quinn: We’ve been exploring generative AI throughout the year for claims. Specifically, we started poking around on document summarization at about the April-May time period, and then over the summer started really going back and forth on building something out. We started that 60-day pilot and that’s the September-October timeframe, and after we have cleared all of our expectations for it, we are currently building the implementation into our production-based claims workflow systems with the target of implementation in January of 2025.
CJ: How do employees feel about the new technology?
Bidinger: From my perspective, I think there’s still an element of kind of the fear of the unknown. That’s why in this effort, change-management, is so important. However, the feedback from those who have had the opportunity to be involved and interact with the tool that we’ve built, and that Andrew’s team has built, has been awesome. So, when you think about how, if you’re a casualty adjuster, a large package comes in, it can sometimes be overwhelming, even for a seasoned adjuster. So what we found with this tool is it was designed to quickly help them understand the exposure and key aspects of an injury while maintaining the decision-making with the individual involved. So, they can really get to some of the detail that they need to properly evaluate a claim in a much quicker manner than what they may have in the past.
Quinn: I get the pleasure of working with employees across the entirety of our organization on generative AI and we have seen the standard technology adoption curve for this technology, whereby we had our innovators who are way out in front of everyone. We’ve probably cleared through our early adopters phase of the folks who are out there, giving it a try now. This pilot group is probably included in that early adopters phase and we’re likely transitioning here into the early majority in 2025, where most folks get a handle on generative AI, utilize generative AI give it a try and what you start to hear from folks is they often will transition from the skeptical into the excited as they realize what generative AI is and what it isn’t. I think there’s often a lot of fear when you have not yet used it. I’m thinking of some of the abstract possibilities, that once folks are getting their hands on their tools, seeing what it can do and what it cannot do, we’ve heard a lot of excitement about how it’s going to improve their own work excitement about how it’s going to be able to build upon their knowledge and expertise, and there’s a lot of optimism within our organization as we think about utilizing this technology moving forward.
CJ: How do you feel about it?
Bidinger: I think of the excitement and it’s a kind of a window into the future, I feel like is a good way to put from my perspective. Many carriers, we’ve probably done things relatively the same for quite a long period of time, and I think some of the generative AI solutions or capabilities almost open up innovation on a broader scale in the organization. I think as people start to use the tools that we’re building, I think they see the potential of ‘OK, how could I apply this elsewhere in my role at you know at Westfield or within the industry,’ and I think that leads to a lot of innovative ideas not just from the strategy team or from our IT area or from our data scientists, but from the from the frontline claim handling team.
Quinn: I recognize that I’m a biased source as I work with this technology day-in and day-out, but I’ve always come at this technology and others from both an optimist and a realist perspective. So, I’m incredibly optimistic about this technology because I see a lot of possibilities for how this technology can make folks’ lives better, how this technology can help our policyholders. I’m incredibly excited about those possibilities and I’m also incredibly passionate about building this technology out as a normalized way of doing work in our future. So, I see this technology as something that is going to significantly change and alter how we do work, and that excites me because the future that I’ve seen from working with this technology is one where we are able to more effectively more efficiently be able to complete our jobs.
CJ: How do you or will you measure the success of this technology?
Bidinger: We are partnering with our with our data team as well as some folks we have in our quality assurance team. Right now, we’re in the ‘We don’t know what we don’t know yet’ phase, and we’re continuing to evaluate what metrics what KPI’s we want to assess long term. Down the road, I think there is the really good possibility that we could see potential positive impact on accuracy of our claims. So when you think about a 1,000-page demand package or medical records, the person going through that, there’s the potential that they could miss some things. Having the technology as an assistant along their side really has the potential to improve the speed at which they can get to an accurate answer. So, there is the potential that we could see positive impacts on that on that accuracy number as well as probably some other KPI’s as well.
Quinn: Within my role, I spent quite a bit of time thinking about measurement and quantification, but I do think less about measuring the success of the technology and rather I think a lot about measuring and quantifying the business outcomes from the application of that technology. So, what that looks like is quite varied. We are often seeking to measure the accuracy of the technology or the reliability of the technology or the quality of the information that’s coming back from our utilization of generative AI, and then we are measuring and thinking about how it impacts our teams. Are our teams able to save time? Are our teams able to handle claims more accurately? Are our teams able to have different results from the application of this technology? So, as I think about this, I’m thinking quite a bit about how we’re measuring the outcomes of the application of that technology rather than just measuring the technology itself.