The Robots Took the Easy Cases — What's Left for Human Underwriters

Paul Tyler (00:02) Hi, this is Paul Tyler, and welcome to another great episode of the L&A Hub podcast. Today we're talking about a topic that matters to anybody who's been in the insurance business — underwriting. And we're going to talk about AI with a really interesting subject matter expert, because so many of these conversations about pricing, underwriting, and AI lead from the carrier to the reinsurer. Today we have Frank Chechel, a good partner of ours who's also the Head of Underwriting for Gen Re here in the US. Frank, welcome to our show.

Frank Chechel (00:47) Thank you, Paul. It's great to be here. This should be fun. It's interesting you mentioned some of those things, because a little bit about my background — I'm an actuary. And that's interesting right off the bat, because I'm an actuary leading our individual life underwriting team.

I worked on the direct side for a number of years, early on in product development — IUL, UL, VUL, term, all the different riders. Did that at a few different carriers, had a great time. Then about ten years ago, one of my bosses asked me to lead a new initiative called Underwriting R&D and build out a team. I knew just enough to be dangerous around underwriting, but once I dug into it — accelerated underwriting, new data sources, automated underwriting, changing the purchase experience for the customer — I really enjoyed it. That's how I fell into underwriting, first on the direct side and now on the reinsurance side.

And in my current role I get to lead an operational underwriting team, too. I have twenty underwriters who do facultative underwriting supporting our clients — looking at impaired risks, looking at large cases, helping them get to a decision. It's been a fun journey.

Paul Tyler (02:28) It's interesting — reinsurers tend to be a magnet for exactly the kind of experience you're bringing to the table. But a lot of people don't actually understand the connection between underwriters, reinsurers, and pricing. Let me frame it this way: why can't I get a discount for wearing an Oura ring, having great results, and being able to show you I'm five years younger than my actual age?

Frank Chechel (03:07) It's funny you mention that, because to me that could be the future. This is the exciting thing about underwriting — we've been doing it for a hundred years, and I think we're just scratching the surface.

Think about the process. For the last hundred years, if Paul Tyler comes in and applies for insurance, we're going to poke you and prod you and ask for all this information. We're going to learn everything we can about your health, then give you insurance. And after that, we basically hope you live a long time — but we're not really exchanging information anymore. The internet, network devices, Oura rings — they change that paradigm. Suddenly there's the possibility of ongoing information exchange, and that reshapes how we think about the business going forward.

We're not there yet. But it's been really interesting to watch the evolution, where for certain people you're collecting less data — you don't need to poke and prod them. These are the questions we're wrestling with from an underwriting perspective.

Paul Tyler (04:35) A lot of change gets driven by external events. I've never seen the study, but I'm sure someone's done it — COVID felt transformational for underwriting. At the company I was at, you either made the investment to underwrite based on data, or you were basically out of the business for six or twelve months. It drove an enormous amount of investment, which then slowed down. And now — challenge me on this — it feels like AI is another acceleration. I remember early ITC events in 2017, 2018 with machine learning models, and everyone said, "That's interesting, it can read stuff." Now the adoption curve on AI in underwriting feels steep again. Is that a fair picture?

Frank Chechel (05:42) I love that observation, and I think it's spot on. As someone who was around before COVID, during it, and after — that's exactly right.

What was interesting during COVID: all these things we'd wanted to experiment with — "hmm, this might help the underwriting process, maybe not as good as the insurance labs we like to collect, but it could help" — suddenly the pandemic hits and it's a real accelerant. As the R&D guy trying to get funding for experiments, all of a sudden everybody's saying, "Turn on all the tools and let's see what we see." I don't mean to make light of a pandemic, but it drove an explosion of underwriting innovation, understanding there'd be some mortality slippage. And the demand for insurance was remarkable — people confronting their own mortality, wanting to protect their families, and that carried on well beyond COVID.

One other thing. Pre-COVID, we all knew John Hancock's Vitality program — the most famous example of wellness incentives, where good behaviors earn premium discounts. A number of carriers were experimenting with that. But during COVID it became a nice-to-do, not a have-to-do, and wellness innovation dried up a bit. Now, as mortality trends have flattened, customers are leaning back into wellness. Think about it — as the carrier, the reinsurer, and the client, we all want you to live a long time. There's a positive ROI there. So can we give you tools to live a healthier life? That benefits everyone. Wellness is definitely a theme we're seeing come back.

Paul Tyler (08:33) I've seen it from a lot of directions — group benefits, the EAP side, the employee fitness side. We're living in a really interesting time. Now, on AI: back in 2018, 2019, companies were doing basic summarization of data and people were skeptical. I think we're all past skepticism now. But if you look at the layers of AI applications today — from "I've got a set of medical records I can summarize" all the way to, at the extreme, an agentic underwriting department — where are we on that continuum?

Frank Chechel (09:28) I recently came across an article on this — it was from RGA, one of our competitors, but I'll give them a shout-out. They break AI into types.

First, predictive AI — a predictive model. You take historical data, build a model, make decisions going forward. That's an area where the industry has made a lot of progress. Some of the investments we made during COVID — better systems, better API connections, bringing model scores into automated decisioning — are paying off.

Second, generative AI — summarizing pages and pages of electronic health records or APSs into clean information. We're seeing progress, but here's the challenge: with summaries, like most AI use cases, if you get 90 or 95% accuracy, that's amazing — it only hallucinated one out of ten times. In underwriting, let me tell you, that's a good way to get fired. You have to be accurate 95, 98% of the time. It seems like an obvious use case, but it has to be really tuned and highly accurate. We're making good progress closing that gap.

Third, agentic AI — where the system runs multiple steps of the underwriting process. That's where we're just scratching the surface. I've read a lot about the "jagged capabilities" of AI — it's good at certain things and not others. I like that framework because it lets you map where we actually are across each layer.

Paul Tyler (11:36) I think of these in layers too. Document summarization saves time — but I've still got to check it, because it has to be right. On the agentic side, the biggest power I perceive — tell me if I'm wrong — is workflow. Something that goes in, sees where a case is, sees its status, optimizes the workflow. How soon does that become table stakes for underwriting departments?

Frank Chechel (12:27) That one's trickier. Come back to predictive AI for a second — those models don't just ask "what class does this person belong in," they also ask "do we need to order additional requirements?" And a big part of the innovation people forget about is the data coming in. Ten years ago, Rx and pharmacy scans were the new thing — we're still learning how to use them. Then you layer in medical claims data, and now you have more data available, but the Rx says one thing, the diagnosis data says another, the application answers say a third. You almost have more data and more contradiction. So how do you build models to handle those situations and the complexity around them? I think we're still more in that predictive-model role. We've got a few steps to go before the agentic piece.

Paul Tyler (13:48) Let me keep going. MCPs — the game-changer for me was our IT department finally turning on the connection between Claude and Outlook, and Slack. Do you have Slack at Gen Re?

Frank Chechel (14:15) We don't. We have Teams — which is like poor man's Slack. Same thing but worse. I have fewer GIFs to use.

Paul Tyler (14:29) Do you see the future this way — and tell me if I state it wrong. You've worked in different parts of these companies. We've used Gen Re's model, RGA's, Swiss Re's at one point. These were manuals first. There's been some shift toward data, but when will I be able to click Claude and say, "integrate with the Gen Re manual and let me talk to it"?

Frank Chechel (14:56) I think we're getting closer. I've seen companies trying to get there — ingesting the manual, building that persistence of knowledge, holding onto it and making more decisions from it. People are leaning in.

But here's what's interesting: more and more of the simple cases have already been automated away. What's left is the gray zone. We've been debating a lot of alcohol-related cases lately. You get three, four, five pieces of information, and if you put that case in front of five different underwriters, you very likely could get five different answers — standard, slightly substandard, significantly substandard, decline. That's hard. And that gets at the question you raised before we started, about whether the robots just take over or humans stay involved. My leaning is that the human job changes, but there's absolutely still a need for humans. I even see underwriting as a growth opportunity. What you do as an underwriter will keep evolving.

Paul Tyler (16:37) Let me pull on that gray-zone thread. You've got the manual, the processes, then the underwriter making the call — and maybe a culture where you know if you draw this underwriter, good luck, versus that one. If you could fine-tune an open-source model on top of the manual to get a consistent decision-making process — or at least know how wide the bands are — have you started to explore that?

Frank Chechel (17:33) That's one of the areas we're looking at. Where are the less-messy cases? Alcohol is messy, but are there impairments at the ends of the spectrum — clear declines — that we can get out of the process? That's what I see happening in more underwriting shops. And here's what's interesting for the human underwriters: there are no easy cases anymore. Every one is tricky, takes time, has confounding information. The easy declines and the easy preferred risks — the robots are handling those. So the human's skill is in the difficult ones. There will always be space for the people who can understand that gray zone, take all that confounding information, and make a strong, risk-based decision.

Paul Tyler (18:38) Talk to me about the skills. If I were heading into underwriting, what's going to propel my career? And what does the job look like in 2015 versus 2035 — or maybe 2030, since it's not even that far away?

Frank Chechel (18:59) There are various paths, and it's definitely evolving. Some people come in with a bit of a medical background — nurses and others who segued into underwriting. However you get in, you start by underwriting cases. Ten years ago, when I was building the R&D team, I was convincing line underwriters to try new things — "this may not work, but let's try it." Now you've got people who are genuinely data-fluent.

I'll give a shout-out to Matthew Sanford over at RiverSource. I met him and assumed he was a data scientist or an actuary. He's an underwriter — one of the most data-fluent people I've met anywhere. Ten years ago you'd never have imagined an underwriter in that role, but he thrives in it, and there are a lot of others like him. You've also got people on the vendor side building the new tools. It's good for underwriters to get a mix of experiences — maybe rotate into those roles, then come back if they love the production side or want to move into management.

And there's the risk-management side. Healthcare isn't static — what kills you today is different from ten years ago and different from what will kill you in the future. How do our policies adjust when AI invents new cancer treatments? Working alongside actuaries, data scientists, and doctors to figure out that future is another path. That's where I see real growth in underwriting.

Paul Tyler (21:39) It's almost like a new arbitrage of knowledge and experience opening doors. One of the best Medicare supplement agents I ever saw was a nurse — she'd go to the house, gather up all the drugs, and just know what was going on with these people.

Frank Chechel (22:11) She was doing the field underwriting for us. That's great — could we get a few more people like that?

Paul Tyler (22:17) Let's talk about the forward-looking side. When we were pricing products, you had to come in with a credible set of data to justify the price. Here's my question: how and where will AI-enabled devices and services — some of these AI bots trying to provide healthcare, Amazon getting into healthcare — have a material impact on mortality outcomes? And if they do, how do you make that case to a state, or do you just absorb it in a block of business?

Frank Chechel (23:27) We think about that all the time. Even in a new world with AI, we've always dealt with medical innovation and what it means for future mortality. Look at GLP-1s — a lot of folks losing a lot of weight. How do we handle that? If they come off the drugs, what happens? It's still very much emerging.

Every time we do a webinar or presentation on GLP-1s, we think people must be sick of the topic — standing room only. Part of it is personal curiosity, but part is that the science is still evolving. These drugs have been around a while and are generally pretty safe, but what are the side effects as more people take them for different reasons? What happens when they come off? Will there be new drugs to stay on longer term at lower doses?

I think about HIV, too. Part of why we expanded insurance-lab collection was concern around HIV. Nowadays carriers write HIV patients and give them the protection they need — which is amazing. So this discipline of working with the medical doctors to make prudent underwriting decisions on new developments is something we've always done. The pace of change is increasing, so the challenge is managing it — but we want to lean into the processes we already have.

Paul Tyler (25:36) I had Connecticut's insurance commissioner, Josh Hershman, on last week — very forward-looking. Where's the intersection of regulation and AI headed? Some states are putting tight guidelines on AI, others aren't. What's the state of the state on AI and underwriting?

Frank Chechel (26:05) I've generally felt things are moving in a promising direction — more dialogue around different tools. Go back ten years to accelerated underwriting, or even credit-based mortality scores, which we still use. I try to be clear that it's not a credit score, but certain things in your credit are indicative of better or worse mortality. There have been different state-by-state approaches to those tools.

What's been good is the dialogue: what's the actuarial justification for these tools? How do you make sure there isn't bias embedded, and how do you adjust for biases you identify? I've seen a lot more conversation among regulators, carriers, and even vendors going to talk with state regulators to bridge the gap and build understanding. We just want to keep doing that — keep helping each other understand. Generally, it's been moving in a good direction.

Paul Tyler (27:26) I'd absolutely agree. I've seen leading regulators at some of the biggest states take proactive steps to make sure society benefits from AI, which is great. It's just a lot of work to take an idea all the way to production.

Frank Chechel (27:53) Even simple level-setting matters — helping regulators understand that life insurance is a voluntary product, versus health insurance or others that are required. Someone chooses to buy life insurance. So what does it mean to underwrite for a voluntary product, and how do these tools come in to make that process better for people? Those dialogues have been great.

Paul Tyler (28:24) Same question I'd ask at Zinnia: where do you think the most interesting applications for technology will be over the next two or three years in underwriting? In 2018, 2019 I'd have said fluidless underwriting. Where are the gaps today on the technology side?

Frank Chechel (28:49) A couple of things come to mind. First is data. When it comes to AI and building models, most companies hire data scientists and say, "we're going to build models" — but they fall down on the data side. The data structure and infrastructure is a mess, sitting in old, disconnected databases. It's the same thing I remember from my actuarial days — you spend 80% of your time cleaning up data to run your projection models. Same here. So a focus on data management, which isn't sexy, is essential. You can hire the best data scientists, but if you don't have good data management, you're going to be hamstrung. Clients making progress there can then leverage the models and rules engines to build some cool stuff.

The second thing that's exciting — you mentioned having Claude co-work. This may not directly touch the underwriting decision, but I've been very impressed. We've recently started a pilot with Claude, and we're building a new knowledge-management system. We have Source Life, our manual, but our facultative underwriters follow all these additional guidelines that go beyond it to get their jobs done. At most underwriting shops, that's scattered across ten different places — a shared folder, a SharePoint site. So we're asking: can we bring it all together? We've started to mock that up with Claude, and what it can do in short order is pretty impressive. That's an area I'm excited about — figuring out some of the broader problems we have.

Paul Tyler (31:03) The data point resonates. I'm looking more from the marketing side — who people are, where they are. But I've been pushing limits I never would have a couple of years ago. Postgres — a SQL database that's really good at plugging into LLMs. If I don't know the business or the questions I need to ask, just plugging stuff in doesn't get the questions answered. It has to be structured right. I'm looking at how we build relationships inside companies, who we care about. I found you don't actually need that much data — the structure matters more. What's your experience?

Frank Chechel (32:14) On data management, it's interesting. As an actuary, the perfect world would be mortality data going back twenty or thirty years on every insured, with every type of evidence on each one. Then I could model all of it and say, "for these folks I need this piece of data but not that," built by impairment. It would be amazing. In the real world, even if you organize your data, there are still gaps. But at least getting it organized tells you where the gaps are. That's where the art comes in — from the underwriters, data scientists, and actuaries. I have this study, but it didn't include that type of data. I have a separate analysis around insurance labs. How do I combine two studies when I couldn't overlap the individuals across them, to make a decision on how I'll underwrite and what mortality I'll get?

So it's getting your data into as good a shape as you can, then accepting that no matter how good it gets, there will be gaps, and you'll need actuarial judgment in real time. That's the fun part, I like to say. I'm a nerd that way.

Paul Tyler (33:54) You can tell I am too. Insurance is one of those businesses where you think, "I'll do this for a couple of years," and then you look at the clock and it's been a long time. Frank, this has been great. If people want to learn more about what you're doing and how Gen Re could help them build their business, what's the best way to connect with you?

Frank Chechel (34:21) Easiest is to find me on LinkedIn and send me a message, or just email frank.chechel@genre.com. I'm usually pretty findable. It's been a great journey — I've been at Gen Re three years, and we've been growing and building out our business. You guys have been great support to us; we use the TPP system for our facultative underwriting, so we appreciate everything you've given us. We're expanding what we do for our clients, so if anybody wants to catch up on any of the topics we discussed, I'm happy to.

Actually — I have a question for you before we wrap up. You're the podmaster here, but I'm curious: talking with all these different people, what's been the biggest aha for you? Where do the opportunities lie, or where's the biggest gap we need to figure out in the life insurance space?

Paul Tyler (35:31) Thank you — nobody ever asks me a question. When I think back to ChatGPT 3.5 — where are we now? I was blown away. I remember the internet, the website; this is probably going to be bigger.

Frank Chechel (36:02) I still remember getting my first BlackBerry and being able to render ESPN.com on that tiny screen — I thought it was the greatest thing ever. This is the next level of that, for sure.

Paul Tyler (36:14) On one hand, think about how much of our business is paper — "where is this document?" There's a lot of basic work you'd think this could rip through. On the other hand, this is a complicated business. I've seen so many people come through saying, "I'm going to disrupt this thing," and there are real reasons it's painful to change here. I saw the opportunity and thought, this is going to take forever.

What surprises me, Frank, is how uneven the application of this stuff is. Some companies are just struggling — no surprise. But then some you'd expect to be way out front aren't, and some small ones are doing genuinely innovative work. So my biggest surprise is the unevenness. It depends a lot on having a leader who prioritizes it — but also on having some DNA inside the company, where the people two and three levels down are curious and open to experimentation. And I'll tell you: the arguments I used to hear about "what's the ROI" have shifted to "how much will it cost and how fast can we do it." My token costs have gone through the roof.

Frank Chechel (38:01) That's right. And on complexity — think about distribution. A number of companies tried to distribute direct and found out that selling insurance is hard. Any successful or unsuccessful life insurance producer will tell you that. It's sold, not bought. People do all the research online, but you still have to hand-hold them through the underwriting process to the finish line.

When we first introduced accelerated underwriting, we told producers, "We're going to reimagine your process — 30% of your clients won't need an insurance lab, isn't that great?" And they said, "No — you're messing me up. The way I work is I have an opening meeting to get the client comfortable with the underwriting process, then a second meeting to go through the offer." Enough of them got comfortable eventually, but you had to engage them: "What if I could get you an offer right in that first meeting? How would that change things?" Some figured it out. But understanding all those stakeholders is where the complexity comes in. Thinking through every stakeholder and the impact on them is key to getting adoption and buy-in. That's part of the fun for me.

Paul Tyler (40:11) It reminds me — when I first got into the business at MetLife, I got to know some of the top reps. One of them sold insurance by saying, "Apply, and you'll get a free physical."

Frank Chechel (40:25) I was going to mention that. Shout-out to a good colleague, Dominique Baud, an actuary now at Prosperity Life. That was his line — underwriting is a free physical. Everybody else in the room said, "No, I don't think people think about it that way," but it's true. I'll have to send Dominique this podcast.

Paul Tyler (40:50) These complicated businesses — the motivations, how people adapt to processes — it's counterintuitive.

Frank Chechel (41:02) Exactly. To bring it full circle to wellness: we know quite a bit about your health by the time we've underwritten you. Often we just say, "You're preferred," or "preferred best, here's your premium." Could we give folks a little more? "Here's what we learned about your health and what you could do to get better." That's something companies are starting to lean into.

Paul Tyler (41:30) I think that's the future. It's a service, not a piece of paper.

Frank Chechel (41:36) Yep.

Paul Tyler (41:36) All right — we're well over your time. Thank you for hanging out.

Frank Chechel (41:40) This was a fun time. I enjoyed it, Paul. Hope we can do it again sometime.

Paul Tyler (41:47) Thanks to all our listeners — join us again next week for another great episode of the L&A Hub podcast.