AI in Insurance Claims: How Ascot Started with FNOL Automation

Introductions: Ascot and Bevaya

Jermaine Everett, BevayaAll right. Thank you guys for taking the time out here with us today. We still have some people coming in, but we're going to take about 30 minutes to just talk about AI and all AI and what we're doing here today with Bevaya, but most importantly, what we're doing with Ascot. And I have Marina, who's here with me, the most important person on this stage, of course.

Marina, do you mind just introducing yourself to the team?

Marina Barg, AscotSure. Everybody can hear me? I can't see anyone. I feel like I need shades for this. Marina Barg. I'm Chief Claims Officer of Ascot US. For those of you who don't know Ascot, we're a specialty insurance company located in three geographies, US, UK, and Bermuda, and doing all things specialty, so exotic lines like ocean and marine, D&O environmental, casualty, and the like.

Jermaine Everett, BevayaAll right, perfect. And so with that, again, my name is Jermaine Everett. I oversee sales with Bevaya. To give a little bit of background for those that have been in the space for a while, we've been around for about eight years. 

And we're really excited because we just launched our Bevaya platform, hence the name change. And so with us, we have about 120 plus AI agents actively in production at scale across commercial P&C with many, many customers, both carriers, some of the top carriers in the world, some of the top TPAs, as well as brokers. And with us, you'll hear us talk a little bit about InsurGPT, which is our mosaic of models that we use that power all of our AI agents as well. For us, we actively focus about the full value chain in insurance, so that's both in policy servicing, things like making changes to your policies, doing things like premium audit, end-to-end workflows, and endorsement processing.

We also focus on things like underwriting, your underwriting submissions that come in, where you're looking at a massive amount of data that's coming in and various carriers are trying to solve for understanding that information, comparing it to your underwriting guidelines and actually quoting that business faster.

Specialty insurance claims and AI partner fit

Jermaine Everett, BevayaBut for today, we're going to focus on claims and the claims experience, and in particular, the work that we're doing with you guys, where you started in FNOL. I know we're talking about some other areas that we'll be expanding with as well. But I don't know if you recall this, but our first meeting that we had, large group on your side, we started telling you about all the things that we're doing with AI agents and FNOL. You stopped me right in the middle and you said, "Jermaine, but do you do specialty? I don't want to hear about anything else."

And thankfully, the answer is yes. But speak to the audience here that's not familiar with specialty lines, why that was such a tipping point for partnering with somebody that understood specialty, and what challenges do you face in that space?

Marina Barg, AscotYeah, thank you. And just one more add to that. We met you and we found out about Bevaya and at that point, Roots AI through ITC. So big plug for the conference and the assistance that it gives carriers in trying to figure out what the next upcoming technology transformative company is.

So yeah, we are Ascot, as I mentioned, we are a specialty company, so we think of ourselves as shiny, exotic, very different type of work that we handle through the policies that we write. So I mentioned some of the lines that we do. We do everything from environmental and the risk associated with that. Inland and ocean marine and marine claims are very much in the news these days and with the Gulf Coast situation or the Gulf situation.

We do space insurance, so satellites that go up in space and the launch that's involved. We just launched kidnap and ransom. So not the usual stuff, not the stuff that is generated or generates a lot of volume. We're more of a severity company. And to that end, we really need to partner with vendors and we don't really call them vendors.

We do call them partner because once we get involved in a relationship, it's sticky with us because we want to make sure we partner with people who understand specialty claims and the nuances of it. And don't give us an out-of-the-box approach that's not really relevant for the lines of business that we do.

Ascot's AI center of excellence

Jermaine Everett, BevayaAnd keep me honest, this was your first AI deployment or your first jump into the AI space?

Marina Barg, AscotIt was. Can you please say again? Yeah. We were, I feel, a couple of years ahead of maybe most insurance companies out there. About three to four years ago, before AI became the only topic that's talked about in any conference. Outside of the industry, we hired a gentleman who's probably here today, Christian Toft Nielsen.

Yes. Who has been doing AI for the last 30 years, who knew that AI existed that long ago. He comes from IBM and has a wealth of insurance, and we brought on Christian to start an AI center of excellence, knowing that AI is going to be a big part of what we do, given the volume that we take in and kind of the unstructured data that we deal with on a regular basis.

Where carriers are with AI adoption

Jermaine Everett, BevayaSo just maybe really quickly, taking a step back to where we were about a year ago and just getting a pulse in the room. At that point, you're looking to go into your first AI deployment. You're spending a lot of time actually thinking about it and where do we get started. For just a quick show of hands, who's actually in production using AI at any part of the company that you're representing today?

I see. Okay. How many are still in the idea phase or kind of evaluation of how do they do it? Where do they focus that? One, two. Okay. All the Bevaya people. Let's go to the-

Marina Barg, AscotThe rest of them.

Jermaine Everett, BevayaYep. There's the prospects. But it's an important journey. And I don't know necessarily how in your seat you navigate all of the noise that's out here. Some of this is should you try to build it yourself? Some of it is obviously there's so many companies talking about the same things.

How did you go about weeding out the noise a bit, landing on the best approach? And then maybe the second part to that, we focused on FNOL. Why was that the starting point?

Marina Barg, AscotYeah.

Choosing an insurance AI partner

Marina Barg, AscotWe were lucky at Ascot from a claims perspective in that our Ascot technology transformation actually started with claims, which is really unusual. It didn't start with underwriting. It didn't start with billing. The focus was on claims because the company recognized that claims is a service that is actually sold by our underwriters, regardless of what they think, because once the policy is bound, the question that we most typically get from our insureds is who's going to be handling our claims and how is that process going to work?

So the company recognized the importance of claims in the value chain of what we provide to our clients. And we launched with a very well-known entity on our claims side, who it prominently features outside on the floor and knew that that was going to be just the beginning. So after we launched with Guidewire, our next thought was, what can we build on top of that? And what can make us more efficient?

And you're right, there's quite a bit of noise out there, and you never know who to believe, who to trust, what's the substance, what's the hype? Will it actually work? Will it work for a specialty company? Hence my question to you in our first meeting. And it was really important for us to find a partner who grew up in insurance, who wasn't necessarily doing banking one day, pharma next day, and all of a sudden decided to dabble in insurance because that was the flavor du jour.

So we found you guys here and we knew that you had deep insurance roots and could help us not only in the implementation process but the thought process of which way should we go with any project that we start.

Build vs. buy AI for insurance claims

Marina Barg, AscotAnd having Christian on board and having a center of excellence for AI, it's always a question that we ask ourselves, do we build? Do we buy? And there are a lot of considerations that go into that, because whenever you build something yourself, you have to not just build it, but you have to support it and innovate in that regard and keep on top of the most current technology. And sometimes that's the way to go, and sometimes we're better off with partnering with experts in the field who know what they're doing.

Why start with FNOL and review processes first

Jermaine Everett, BevayaPerfect. And so with that, for everybody or anybody that has deployed any AI solution or trying to replicate what your team is already doing manually, there's obviously a level of depth that you're usually not going to get into when you're just having your day-to-day processes run. It requires you to scope your work, understand your own business rules, and really, truly evaluate if we're doing business the right way because you essentially have to have your AI agent doing all of those things. And so we went through a deep exercise of understanding the way your team does the work.

Was some of that, though, enlightening or provided some visibility into maybe some inefficiencies or just the way that we're doing business? Is this the right way to do it? Can you talk to me a little bit about the onboarding process for you?

Marina Barg, AscotYeah. We knew about the inefficiency part of it, and we started with FNOL because that's kind of the part that made the most sense, and quite honestly, it's one of the ways that we can test you as a partner. Something that works already with us, so if all things fail, we could always go back to our existing process, and luckily it didn't fail. It succeeded very much.

But part of, I think, any thought about onboarding AI is going through your existing processes and trying not to automate things that have always been done a certain way. It's a great opportunity to review the processes and to figure out why are we doing it a certain way. Because it really doesn't make sense to automate a process that doesn't work or a process that requires a lot of workarounds. And it was good to have a thought partner in that regard who would question how we currently do things and try to offer solutions that could be better to serve our needs.

FNOL automation results

Jermaine Everett, BevayaSo as we kind of go into just some of the results here, and for those that don't know, when you present at ITC, they say, "Give us our slides like a month in advance." So there's updated data here, but I believe when we put this together, we had about an 85% no-touch rate, 99% success rate. But I think today we're about 90, 91% right there.

Marina Barg, AscotAnd going higher.

Jermaine Everett, BevayaAnd going higher. I know this is the foundation for you. This is where we're starting. What's next for you? What are you guys focusing on in 2027, 2028? Where do you see the scalability of AI going?

How AI automates FNOL from a claims inbox

Marina Barg, AscotSo let's start with FNOL in general and the way that process worked and what we can do to improve that. For a specialty company such as ours where, as I mentioned, it's not about volume even though we have a bit of that, but it's more about severity and it's working with commercial insureds. We don't have call centers. We don't have large amount of people who are thrown at the FNOL process.

We have a very lean group that manages our FNOL through an email box. Right or wrong, that's how we started. So we have a dedicated email box where most of our FNOL go to. And the way FNOL looks in a specialty company, it could be a three-line note from a broker saying, "This is the policy number. We have a claim with a one-line description. Please set it up."

Aside from that, in that mailbox, just because people on the brokerage side, on the client side are used to using it, there's a ton of other information that gets sent to that mailbox. So anytime there's an update on a claim, there are documents that go in there. There's inquiries about claim status. Anything and everything that you can think of because that's the only email box that external world knows goes through there.

So our challenge was to try to figure out what's truly FNOL and then to try to set up ways in which we can manage the rest of it in a more efficient way short of throwing bodies at it. So we worked with you really carefully to build rules to have various calls to our policy side to ensure that we're connecting the FNOL to the right policy and then take it a step further and actually setting up the claim in our Guidewire in a way that used to be done by people. And really cut down our process time, which helped not just us on the processing side but really helped our insureds because they could get their claims set up a whole lot faster, acknowledged a lot faster, triaged to the right adjuster so that you start working on a claim a lot faster. And it creates a better experience not just for us internally but externally as well.

The claims value chain

Jermaine Everett, BevayaAnd so for the claims value chain, essentially a new loss comes in the door.

Marina Barg, AscotYep.

Jermaine Everett, BevayaThere's then supporting documentation that comes in on that claim for the handling purposes. There's evaluation of is this claim payable and how much should it be paid and what are the coverages for this claim? There's invoices that need to be paid. There's adjuster review of the documents that come in the door.

When we sit with our customers, we try to get them to really think about the full value chain of a document or a piece of correspondence that comes in the door. On average, if one thing comes in the door, usually you have five or six people that need to look at that one correspondence, right? For different reasons and for different purposes. So the costs continue to compound on that one piece of information.

Claims indexing for supporting documents

Jermaine Everett, BevayaWe are now getting started at looking at claims indexing and evaluating all that supporting unstructured documents. As you stated, the FNOL box gets a lot of things that are not related to a new loss and just people sending their work in. Why is that the next step for you as you kind of think about just unstructured documents and supporting documents?

Marina Barg, AscotExactly for that reason. So information comes in, there's usually quite a bit of it. It's not enough to just set up the claim and route it to adjuster. We can probably do better.

Jermaine Everett, BevayaYep.

Marina Barg, AscotWhat does that document say, right? That's the summarization part of it. Based on our workflow, what should you do with that document once you identify it as whatever it may be? Escalation internally for stuff that needs to be escalated. On top of it, it's the capture of all the unstructured data.

Then that can be mined from underwriting perspective, from actuarial perspective, from analytics perspective. So that would be our next step. To index that, meaning not to throw more bodies at additional correspondence that comes in, be able to tie any of that to an existing claim, kick out anything that requires human intervention, and then actually do the extra step of identifying what's in that document and how does that affect the lifetime or the life cycle of a claim.

Change management for AI in claims

Jermaine Everett, BevayaSo AI, especially for most engagements, usually the directives are coming from the top down. But for a successful deployment, we need your people to buy in. They need to support. They need to train the AI agents.

Marina Barg, AscotYeah.

Jermaine Everett, BevayaThey also have to understand the vision to the most that you can kind of explain that. But that can be a scary proposition if somebody's bringing AI to do work that you're also doing or to help you do that job. There's a change management exercise of that. How did you handle that?

How did you get the messaging out to your team as far as where you're planning to go and any potential impacts on them?

Marina Barg, AscotYeah, and it's something that surprised me and actually goes back to our implementation of Guidewire where we started off with a system pre-Guidewire that really wasn't suited for purpose. And we managed to deal with it and we managed to hobble along until a decision was made to enhance our claims capabilities. And you would think that all the adjusters on our side would be screaming and yelling for the new shiny system that helps their life easier. And I was told early on that you're going to need a change management approach to this.

And my thought was, "No I don't." Like, "We're all in. We want change management. We asked for this. We got it. We got senior management to buy into the fact that claims is the most important part of specialty insurance. I don't need claims management." You'd be surprised.

You get people who are so set in their ways for whom changing minor aspects of their workflow just creates, you know, "No, I've never done it this way." So we knew going into anything after PolicyCenter that we need to take care of change management, ensure we bring people on board. So much like with our original implementation of Guidewire, this was really a claims and business-centered and business-led initiative. It wasn't IT telling us what to do.

It was our own claims people working through the process, working with you to try to figure out what works best for us. And from that point, change management became a little easier. And once people see the results and can look past the initial ramp-up and the training of the model, and see the results and see the output, it was easy to get them to come along.

Helping claims professionals decide faster

Jermaine Everett, BevayaSo when we made our change to Bevaya, it was centered around our new platform, which was really focused on giving IT builders a tool that they can solve solutions that you needed through your system. I use a line where I say that for every CEO that has a directive to say onboard AI, there's an IT leader that also has a directive to say build solutions for the company to solve our problems. Obviously for this, you're working with us, but there is still going to be an investment internally to build solutions and solve for business problems as well. What is your internal team really focused on trying to solve for, versus we know that there's other things that we're doing together, but where does that focus as you guys go into next year on internal solutions?

Marina Barg, AscotI think ultimately we want to get to a point where people, and you can't leave people out of the mix, especially in specialty lines that we handle. So we're looking for solutions that will get our people to be able to make the right decisions faster. That's ultimately, to me, the purpose of AI. We're not doing it so that we can check off the box and say we've done AI because the CEO is expecting it or the board is expecting it.

It's really to make our people focused on the things that we've hired them to do. And we typically hire experienced folks, professionals who have degrees outside of claims management in their specific areas of expertise, and we want them focused on making those decisions better, faster, more efficiently.

Setting accuracy expectations for AI

Jermaine Everett, BevayaWhen you guys started into this, what we tend to see in our deployments is the level of expectation for AI is near perfection, right? If we have one error, throw it away. We can't use this. There's no way we could use this. But if we were to look at our historical performance in the same lens, we would realize that AI, maybe it's 98% accurate, but we were 85% accurate, or we don't necessarily know what our success criteria is going into an AI deployment.

Do you think that was a challenge for you guys to have either level set what your expectations are, and maybe you thought one thing starting it, but you realized as you kind of got educated on what AI is and is not. Do you think that was clear going into it, and where do you think you got that realization if it wasn't clear?

Marina Barg, AscotYeah, I think you're right. We expected to turn on the button once everything was done and things load perfectly, and they didn't, for reasons that probably didn't have anything to do with you or us. Maybe it's the underlying process. Maybe it's something that we didn't think of.

And we knew that it was going to be a journey, so we focused on the fact that every step of the way needs to be an improvement over last step, and that's where the business collaboration really came into full view. Because it wasn't you left to your own devices trying to figure out how to improve that rate. It was us communicating on a regular basis saying, "We can do this different. We can accept perhaps a little less here because it's not as important to us."

We launched in February. And we tracked the success rate as we keep going forward. We knew that it was going to be at least a six months process for us to train the model in a way where we have confidence in it. We had hiccups along the way that we handled, and I guarantee you, there are no IT project that goes forward that doesn't have those hiccups.

And we worked through them.

Jermaine Everett, BevayaSo just to recap, you said the deployment was easy, perfect, and we clicked the button and everything went-

Marina Barg, AscotTotally.

Jermaine Everett, BevayaThat's what I heard.

Marina Barg, AscotIs that what you heard?

Jermaine Everett, BevayaMy wife is working on my listening skills, but I think that's what I heard from you. And so with that, I know we have some time, and I do want to allow anybody in the audience to ask questions. So feel free to raise your hand if you do have any for Marina or myself.

Will AI handle insurance claims end to end?

Jermaine Everett, BevayaBut, as you guys think, I told you what we'll do, a true false and why game, right? And so just two quick questions and just give me your thoughts on it. Okay?

Marina Barg, AscotMm-hmm.

Jermaine Everett, BevayaThere will be a near future where AI will handle claims from end-to-end decision-making and all.

Marina Barg, AscotFalse.

Jermaine Everett, BevayaWhy?

Marina Barg, AscotFirst of all, the regulators will never allow it, just a pure fundamental reason. But there are nuances to what we do. There's nuances to coverage analysis. I think that a claims adjuster is the hardest job in the industry. You wake up every morning knowing that you're going to have conflict somewhere along the way in your day, and you need to be prepared for that, and you need to have the stamina and the personality and the desire to be able to handle that.

There are clients, there are brokers, there are insureds who will see things differently from you. There's always a piece of judgment in the decisions that we make. And I don't think you can train an AI model, no matter how long you train it, to have that level of understanding and expertise in the lines of business that we write.

Audience Q&A: FNOL implementation challenges

Jermaine Everett, BevayaOkay. Any questions for... Yes, sir.

Audience memberCan you talk to us about some of the challenges that you may have experienced along the way in the implementation process and the rules in policy?

Marina Barg, AscotYeah. Some of it was our internal plumbing, being able to connect Bevaya to our policy side, right, and be able to pull the right policy for the right date, and to be able to get into Guidewire the elements that they need in order to be able to set up a claim. So that was hard. The other piece of it is just getting the rules established in terms of what constitutes an AI, sorry, an FNOL, and treating it as such because there is a lot of other information that comes through that mailbox.

The other aspect is just that, but everything else that comes through that mailbox, and ensuring that it goes to the right entity in the right time and allowing us a feedback loop back to us to knowing that something was done with it. Because that's when you get into regulatory problems, right? When you tell them that AI was handling this and there's no view into what happens with any notice of loss that comes through to us and then eventually gets handled. If we can't explain it, if we can't provide trace and audit record of what happens there, we're going to have a problem.

So I would say it's the internal components of any company and ensuring that your policy side is ready to accept that challenge and be able to provide information back to you ensuring that it's accurate.

Jermaine Everett, BevayaAnd just to give a little bit of color from our perspective as well, when you hear us talk about deployments or implementation timelines, you'll hear us reference timelines can be as short as three to four weeks. And what we typically find is, yes, if everything is there and we're ready to go and we have system connections, APIs are enabled, we can go extremely fast to get your AI agent on board, deployed, running in production at scale. But there's just nuances related to your business and your business rules, as you mentioned, looking at why are we doing something a certain way? And you should take that opportunity to optimize your process as well as streamline it.

You shouldn't just automate or build a workflow based upon how you're doing work today because you're probably doing work that way because of certain constraints or how you're just doing your business today. Any other questions as well?

Will insurance regulation catch up with AI?

Jermaine Everett, BevayaOkay. All right. Then I got one more for you.

Marina Barg, AscotGo.

Jermaine Everett, BevayaI know. We got two minutes. Insurance regulations will eventually catch up with AI.

Marina Barg, AscotFalse.

Jermaine Everett, BevayaWhy?

Marina Barg, AscotWe have 50 states that regulate insurance. There are 50 state regulators who want things done their own way. Technology, I think, always outpaces regulation, as we're seeing now with all the discussions about AI and how and when the frontier model should be regulated. I just think that regulatory approach always trails what's happening on the business side.

And unless they implement some level of AI on their own side, I think it will continue to be that way.

Jermaine Everett, BevayaInsurance regulators implementing AI to monitor the AI-

Marina Barg, AscotYeah.

Jermaine Everett, Bevaya... that's gone rogue.

Marina Barg, AscotExactly.

Jermaine Everett, BevayaGot you. All right. Well, number one, thank you everybody for taking the time here with us. Marina, thank you so much for partnering with us here today and sharing your story and the work that you do.

Marina Barg, AscotThank you.

Jermaine Everett, BevayaAnd for everybody else, thank you as well for taking the time out and we'll end today's session. Thank you so much.

Marina Barg, AscotThank you.