Narrator:
This episode of Life Accelerated is brought to you by Equisoft, a leading global provider of end-to-end cloud-based solutions with deep domain expertise in the life insurance industry. To learn more, visit equisoft.com.
Brian Poppe:
If you started 15 years down the road and you're going to ask, is AI part of underwriting? I think the answer is going to be like, yeah, AI is certainly helping the underwriter in some form or fashion. The only question then probably is one, how quickly do we get there? And then two, what does that end state look like? Is it the AI system doing the decision-making for the underwriter or is it recommending something? And then again, how quickly do you get there?
Olivier Lafontaine:
I'm Olivier Lafontaine, and this is Life Accelerated, the podcast for life insurance leaders focused on driving meaningful change through technology, process, and partnership. Today I'm speaking with Brian Poppe, Senior Vice President of Life Insurance Solutions at Mutual of Omaha. Ryan oversees one of the industry's most recognized life insurance businesses and shares a wealth of experience on modernization, underwriting and customer experience. In this episode, we'll look at how Mutual of Omaha approaches innovation while protecting a trusted brand. We also explore the growing role of AI in insurance from underwriting and customer service to virtual agents and operational efficiency.
Brian shares lessons from Mutual of Omaha's AI initiatives and explains why successful adoption starts with improving processes before applying new technology. Finally, we discuss what insurers should be focused on today to prepare for the future of artificial intelligence. Let's get into the episode. Good afternoon, Brian.
It's nice to have you on the show again. I think you've been on the podcast a couple years ago actually. It's nice to have you again.
Brian Poppe:
Yeah, excited to be here and it is nice to be back. It feels familiar and I enjoyed our conversation last time, so I'm looking forward to today.
Olivier Lafontaine:
Good stuff. And one thing that we've done with all of our participants is try to talk a little bit more on the personal side. And so in our preparation, you mentioned that in your free time you do a little bit of DJing. So why don't you talk a little bit about that and what's your go - to set and what kind of music do you play?
Brian Poppe:
Yeah. Oddly enough, I've been DJing longer than I've been in insurance, which is a weird thing to say. One of my friends and I started a DJ business in high school and I just have done it ever since. I never stopped doing it. Boy, if you're letting me pick kind of the music, I tend towards disco or house would be my preferred type of set. Although I got to admit, I've been working on a set recently of, it's yacht rock music. So the think like smooth early '80s with a house beat underneath. I'm tentatively calling it Boathouse. Yacht Rock plus house is boathouse. And boy, it's been a lot of fun to figure out how to make mashups that really work well with Michael McDonald or Christopher Cross. So if you see me somewhere, good chance that I'll be playing one of those.
Olivier Lafontaine:
And so you do sometimes DJ in bars or clubs or events. You do it professionally?
Brian Poppe:
I mean, I do get paid, I guess, so that I suppose makes me a professional at it. But yeah, I'm on the InsureTech circuit. So if any of the listeners are at an InsureTech event, there's a good chance I'll be at a bar or an event afterwards that you can catch me at.
Olivier Lafontaine:
Awesome. So I'll try to find you next time we go to InsureTech Connect or some other event out there. Awesome. So why don't we talk a little bit about your role and what you've been doing here at Mutual of Omaha, what your role is, what your mission is, I suppose, and what you're working on.
Brian Poppe:
All right. So I've been at Mutual of Omaha for 17 years. I currently lead a business unit called Life Insurance Solutions. So that is all of the individually sold life products roll up under that P&L. So I've got everything from marketing all the way through claims, including customer service, IT, all the pieces that go along with that. So we at Mutual of Omaha serve people and not things. So everything that we offer would be somebody that would have a human attached to it at some point along the way. On the life insurance side, we've got, we call it final expense. So the small face amount whole life that folks typically use for a covering a funeral. Term life that everybody's familiar with, I'm sure. And then Universal Life are kind of the three main product categories that we've got. And we go to market in a variety of ways, including direct to consumer, which is kind of unique in the life insurance space.
Olivier Lafontaine:
For sure. Yeah. And you do also health insurance? You said the life insurance side. What other lines of business do you do?
Brian Poppe:
Yeah. So one of the other business unit does... Well, there's two other business units that I'll reference here. One of them is health and annuity. So I used to work with that team relatively closely. And so they've got most things under the umbrella other than major medical. So that's like long-term care, hospital indemnity, a cancer critical illness type policy. And then as I mentioned, they have annuities as well. And then senior health is the third business unit that make up, we call it retail solutions or individual products effectively. Their primary product lines are Medicare Supplement and they also have a dental policy that they sell to a decent number of customers. They just sort of attach the two together. Obviously their target market there in the senior health market is 65 and up, folks who are eligible for Medicare. We've got a couple of other major business units, I guess.
We have a mortgage company. We had owned a bank for a period of time and sold the bank, but kept the mortgage company. We've got, we call it institutional investments. So they're the folks that are out interfacing with some of the very large scale investors who are buying that. And also interesting bit under there is the pension risk transfer for employers who have pensions in their back book. And then finally, the last business unit is our employer business unit. We call them workplace solutions, but they sell policies to groups. And then if you're an employee of that group, you end up getting some sort of coverage, whether it is provided specifically by the employer or you can buy your own and get a little bit more coverage if you need it. So yeah, that's the spread.
Olivier Lafontaine:
Yeah. So in your role, do you address all of those lines of business or only the life insurance or which of those are part of what you are covering?
Brian Poppe:
Under my umbrella is just the life insurance pieces, but you can imagine we all get a chance to work relatively closely together. I've got multiple meetings a week with at least the other individual retail leaders and then at least once a week with the other business unit folks as well.
Olivier Lafontaine:
For sure. And so Mutual of Omaha being sort of a really well-known household brand, how much does that translate into a competitive advantage with regards to the digital first marketplace and direct to consumer? And so do you see that is having a big effect on your success in direct to consumer and online insurance?
Brian Poppe:
Yeah, we sure seem to think so from how that brand resonates with folks. So for those that maybe on the younger side in our listening audience in the late '60s and '70s and actually even up into the early '80s, I guess, Mutual of Omaha sponsored a television show called Mutual of Omaha's Wild Kingdom. At a time when you had maybe four channels to pick from, Wild Kingdom was one that resonated a lot with families. It was effectively Animal Planet before Animal Planet was a thing. And we have millions of viewers. And so now anytime you do any sort of primary customer research where you've got a customer talking about either their products or their experience with insurance, and we mentioned Mutual of Omaha, the over under is generally about eight minutes as to whether or not they mention Wild Kingdom. So you could probably put a line on it and make some money that way.
So it is certainly one that is a very well-known brand with a lot of Americans. And we think that is very helpful to your point in the direct to consumer marketing. It's just a brand that's been around for a while and one that people typically associate with very positive feelings because they remember sitting down as a child, either at their parents' or grandparents' house, whatever, and watching Wild Kingdom with the rest of their family. So it's certainly one that we've used to our advantage and we do think about strategically how we want to use that. We did bring back the Wild Kingdom TV show. You catch it Saturday mornings now in the children's program we block on NBC. Otherwise, if you're a Peacock subscriber, you can find it there as well.
Olivier Lafontaine:
Good stuff. And I guess that does put, do you feel it puts a little bit of pressure that being so public, I suppose, in terms of if you do transformation from a systems perspective, it has to be right. There's a reputation to uphold, I suppose. Is that a bit of pressure sometimes?
Brian Poppe:
I think probably any large company feels that too of like, gosh, when you're at a scale where you're counting customers in probably the millions, you really don't want to screw that up too badly for folks because one, you get a lot of complaints in a hurry. Two, those dollars start to ad up no matter what the quote error might be, whether it is like, gosh, we were down for an hour or two because some tech idea failed, or we launched a brand new initiative and it fell flat on its face. There is some cautionary work that you've got to put in place in order to maintain that brand quality. And you might imagine we've got folks on the brand team who do think pretty deeply about the types of partnerships or the types of sponsorships that we have in order to help protect that brand asset, perhaps more than most as you noted.
Olivier Lafontaine:
Yeah. And every error or problem is magnified in the public eye.
Brian Poppe:
I suppose even more so when you've got direct to consumer as a channel, if you impact producers, they may be okay about sheltering the carrier from the direct customer impact so you can impact them. But once you're out public facing and you're talking to customers directly, things can go south in a hurry because you've just got a lot more eyeballs rather than just the producer community who you certainly want to keep happy, don't get me wrong. But they may be able to do some things that make you look a little bit better than just the website's down or whatever.
Olivier Lafontaine:
When you're direct to consumer, you are compared to Amazon and Google and those guys.
Brian Poppe:
Unfortunately, for those of us in the life insurance market, yeah.
Olivier Lafontaine:
You mentioned earlier, so you have Simplified Issue Life Insurance as one of your core products, I think. What would you say is a secret to that market and how do you do well in that space at scale?
Brian Poppe:
Simplified issue market is a tough one. There were several companies that had attempted to be in the simplified issue business and then ultimately had to back themselves out. My hypothesis here is they couldn't make it work from a profitability standpoint. There's a few things that you have to have working in your favor. One of those is certainly scale, as you mentioned. A second one is even if it is simplified issue, there is still some underwriting questions that you've got to make sure that you get right so you know what is going to go into that risk pool. And then balancing the, we'll quote good risks and bad risks. And in that risk pool is probably the thing that is valuable. Third thing is it is you get a lot of churn from a customer standpoint. The customers, my theory here is that it's easy to buy and so it's easy to probably lapse because you just are like, well, I'll just go through that same application process a month from now.
If I'm squeezed this month for cost or some other opportunity comes my way, it's just like, well, shoot, it was just as easy to get it, so I might as well lose it. So my guess is that the competitors who dropped out in 2024 probably ran into any one or all three of those challenges and decided that it was probably better for them to exit the market rather than the other way around. So scale certainly helps you with getting that risk pool right. If you're in the hundreds or low thousands of policies, you're going to get crushed probably by the mix. Even if you're trying to get that mix right, a couple of claims one way or the other can swing you from profitable to unprofitable. And then your CFO starts asking some questions and you're like, "Well, we're trying to build a market. It's a longer term process than two years or three years or however long you've been in it.
" And I can see that pressure may eventually get to somebody who's trying to manage that, especially if you don't have a good line of sight to how you might be able to get out of there. The other thing that's interesting there is it's like a different distribution base regardless of whether you're going direct to consumer or whether you're going through an independent channel. It's just a different way of you have to manage those producers slightly differently. Messaging is different. Speed obviously matters a ton. Things you might be able to extend out for a couple of days and fully underwritten the fully underwritten market you just can't do in simplified issue because those producers are moving on to the next carrier, to the next application, to the next customer. And so speed really matters in getting that right as well. A lot of things you got to do right in the simplified issue market in order to make it work.
Olivier Lafontaine:
Yeah, because if it's simplified issue, but it's not that simple, then it defeats the purpose of the product a little bit obviously.
Brian Poppe:
Yeah, you've got it. Certainly the speed is the name of the game. That's a lot of reasons why a customer would end up purchasing that product is because you can get them a policy either same day or next day as compared to on the life insurance side at the very least. What might take you 20 something days and a bunch of back and forth with your producer or the carrier who's on the other side of that.
Olivier Lafontaine:
So I guess with scale, the success at scale is actually scale is probably necessary to be successful because to your point, I think you need in order to come up with the right underwriting questions, you do need to ask questions if you want to be price competitive and tweaking those questions to be sufficiently specific to get the right pool of risk and at the same time not slow down the process I guess is a difficult balance and only experience I would imagine. And volume allows you to tweak that to a place where you get it just right, I would imagine. Making it harder for competitors that don't have that experience or that volume to sort of, I guess they come up with generic questions a bit more.
Brian Poppe:
Yeah, I would agree with that. And it's difficult I guess to tell you how you might start one of those things from scratch. It's one of those things that I guess I'm lucky enough to be in my position that we've had folks in the past that have gotten burned by either a unclear question or a poor question or a missing question altogether and then you end up several years down the line with, oh shoot, we were paying a lot more claims than we expected and you got to go reevaluate. Well, we didn't ask about cancer history or smoking history or whatever the answer would be. And now we've got to do that and figure out how to improve that risk pool going forward.
Olivier Lafontaine:
I assume you use third-party data sources or third party saliva tests and maybe LexisNexis risk information and things like that. Does that change over the last couple years? Is that something that is fairly stable or there's more and more data sources that you're able to look at to accelerate the process?
Brian Poppe:
Yeah, I've termed internally the year of 2026 is the year of underwriting. So we're in the middle of doing some underwriting transfer management. The data, as you referred to, is one of those three pillars that we've got in the underwriting space. The other two being process. How can we improve the process overall and then technology supporting both of those things. The thing that we found, I guess it's not necessarily that you're looking for new data sources necessarily. It's just more, can I get the same type of data but in a faster fashion? So underwriting in our case and probably in most of the carriers who are still in existence case, you know to some degree of certainty what type of experience you're going to get based on the underwriting that you've done. So the process here, at least from our standpoint, is not necessarily like do I need to improve the underwriting decision?
But how do I, to the point of speed earlier, how do I compress that timeframe? And then once I start compressing that timeframe, where else can I get that data? So I used to get it from a vendor A and now I'm onto vendor B because vendor B can get me that data faster. At some point the industry might be asking, "Gosh, how do I make better risk decisions?" We haven't necessarily seen that yet. If you follow the insurance industry's trend in maybe starting about 10 years ago, the industry started to do accelerated underwriting programs, which mostly just cut out requirements in the hopes that you'd be able to just get roughly the same mortality mix and not have to order everything at once. I think based on some reinsurance data that has come out relatively recently, the carriers that greatly adopted the accelerated underwriting program are finding that mortality is worse than they had expected, worse than compared to what they had thought.
So the thought being here that like, gosh, there's probably some data that we could get faster. And it may just be existing vendors and you say, "Oh, do you have a new way of delivering this? " rather than, "I'm pretty sure we don't get anything via paper anymore, but via paper that gets mailed to us. Can we get that any faster electronically?" And then make the decision based on that electronic form rather than the paper form that comes in. Those are the types of things that we're doing from a process and a data standpoint. And then as I mentioned, the technology that serves both of them making that decision faster.
Olivier Lafontaine:
Are you thinking of using AI in that underwriting process at all? I know it's been a discussion in all the conferences I've been to in the last year and there's generally speaking obviously hesitation or caution rightfully in adopting AI for underwriting, but what's your take on this? Do you think it's going to go there eventually? Are you looking at that at all or what's your view? To make an underwriting decision or to help in evaluating in a context of simplified decision. Typically it is more rules-based. Obviously you look at certain data and you make a decision, but is there a space you feel is eventually to say, well, maybe we'll have a part of it that's going to look at unstructured data and then AI makes a decision on that and somehow we can control that a little bit from a compliance perspective. And you think there's a future for that or that's far away?
Brian Poppe:
I mean, I guess I don't know how far away it is, but if you look maybe like if you started 15 years down the road and you're going to ask, is AI part of underwriting? I think the answer is going to be like, yeah, AI is certainly helping the underwriter in some form or fashion. The only question then probably is one, how quickly do we get there? And then two, what does that end state look like? Is it the AI system doing the decision-making for the underwriter? Is it recommending something? And then again, how quickly do you get there? I think if you start at the end and then start to think your way backwards, you're like, well, clearly that's going to be the case. How soon do you get there? I guess I've seen more adoption in the property and casualty space and I haven't figured out if that's a function of the regulator or a function of the carrier.
The technology itself is the same maturity. We can access the same AI tools on the life insurance side that somebody on a PNC carrier can. And yet PNC seems to have adopted faster than the life insurance market. So at some base level, if you're saying, well, a regulator isn't willing to adopt this, I can point to property and casualty and say, well, clearly they are. There is a regulator out there that is allowing this. Granted, maybe one of the other major differences between PNC and life is property and casualty has the ability to annually renew rates, whereas life insurance typically is I price once for the next 20 or 30 or 40 or whatever years. And I can change that, but anybody who's changed that pricing typically gets hit with a very large lawsuit relatively quickly. So you are trying to say, well, gosh, I need to probably be a little bit more cautious because I'm locking in this price for the next X number of years, X being longer than probably 10 to 20 years where a property and casualty company can come back a year from now and say, you know what?
We got it wrong. Last year was bad from a profitability standpoint, but we're going to get it right next year. Life insurance companies don't typically have that luxury. But I guess to answer the question, have we though about it? Yeah, in the way that I framed it of what's this look like 10 years from now or 15 years from now? So are there things that we should be doing today in order to prep for that? Data being one of those pieces that we're trying to think about of how do we set up the data so that when that time comes, we're more prepared for it than we are today.
Olivier Lafontaine:
Yeah, that makes sense. And so today, even though it's simplified issue, you still have underwriters that will review. Is it fair to say? Or it's not a fully automated decision. It is a quick decision, but not necessarily free of a human judgment in the case.
Brian Poppe:
Yeah, we've got both. We'd have fully underwritten business as well, I guess, just to be clear. On the simplified issue side, which maybe the point you're making is we do sell more policies there for sure than we do on the fully underwritten side. And it's not everybody gets a instant decision. There are some that get kicked out for a human to review, ask a few more questions of the applicant and then ultimately issue. That issue time is still within two days, but there are some that do not get instant issued or instant rejected. They're reviewed by a human and then ultimately a decision gets made again, typically within 48 hours.
Olivier Lafontaine:
And I'm sure, yeah, you're still trying to do this quickly to preserve the concept of speed and delivering on the promise of the product. Even though they were some kicked out of the promise of the instant issue, you can still get a lot of value by making it simple and quick. And I think that's where maybe eventually we'll see. I don't know if it's a regulator thing because ultimately the regulator will say the ball's in your court to you're accountable as the insurance company anyways. So if you're letting it done, if you do your underwriting using AI, then demonstrate that it's a fair decision, that there's no discrimination and that kind of stuff. So that's what perhaps we'll see that in the future. How do you put those proofs and those programs in place to show that the AI is able to make a decision in a comparable way that a human might do it?
Well,
Brian Poppe:
Would expect that's what a regulator is interested in because that's what they're interested in most of the time of prove to me that this is fair. And then to your point, you're on the hook, so you're kind of beholden to your own underwriting decisions at that point.
Olivier Lafontaine:
So if we move in our preparation, we talked a little bit about how you use AI today, which is not in underwriting. It's more around, from what I understood, call summarization more from a customer service perspective. I though that was super interesting. Do you want to describe that a little bit?
Brian Poppe:
Sure. Yeah. So earlier this year in 2026, we put call summarization in place for using AI. So the AI tool will take the transcript, we'll summarize that and then put that back into the customer service rep's notes. That has, I think, made everybody much happier. The customer service reps generally did not like having to summarize their call, put it into notes. So they're certainly happier with that. Their after call work, which again is certainly less fun than working with a customer, has reduced pretty significantly, I guess, because we've been able to automate that call summarization. I will freely admit, I was probably skeptical at the beginning of, gosh, I think there's so much more we could probably do with AI in the customer service space than just do call summarization. But I was correctly proven wrong that this is a, one, gave us a quick win with a positive ROI and two was relatively easy to implement.
And so you can now build some momentum from a cultural perspective of like, all right, here's how we're implementing AI and here's how it's useful in tasks that we already do day-to-day and have gotten some positive feedback from the employees that are affected. Yeah, so from there we are also moving to a virtual agent. And when this podcast goes live, we should have some portion of our customers that go through that. So I don't feel bad telling you about it because the agent clearly identifies themselves as an AI agent trying to assist you. Goal there I guess is again, to sort of take some of the more rote type calls that come through to that customer service team and just sort of say, gosh, we get calls that look like this all the time. Customer service rep gets tired of taking them because you're again, just having the same, okay, we need to change your address or whatever the thing is.
Olivier Lafontaine:
Or what's the value of my policy? What's the death benefit? Or who's the beneficiary? I'm sure a good portion of calls are always about the same four things, right?
Brian Poppe:
Yeah, you've got it. And so virtual agent will be able to take and handle those calls going forward. So we've piloted that with live customers with live calls. We're learning from that pilot and we'll be scaling later on this year.
Olivier Lafontaine:
Yeah. And presumably that makes it a win-win-win because I would imagine the place you need the people to take the calls I'm sure is something like if they have a claim, somebody passed and they have questions then it becomes super sensitive. I'm sure that's a better use case for people to take the call at that point. And so if they don't have to worry about call summarization or those routine mundane calls, then gives them more time to work on those things and everybody benefits. You still need those humans, the empathy, the customer is better served and the company has better service overall, which presumably translates into better financial results I guess ultimately.
Brian Poppe:
Yeah, you've got it. That's our hypothesis as well and look forward to that one coming true because we think the same thing. This especially, I guess, carries some weight as you start to think about claims experience. And that is probably regardless of whether you're in PNC or life and health. If a customer is calling you about a claims experience, one of the things that we've spent a decent amount of time talking about is do you actually want a human on the back end of that because of the empathy that you're referring to? It seems like even if your AI agent can express empathy, there's probably something to that little bit of magic of having a human on the back end of that who can share some of that pain with you and at least understand and then be willing to help in whatever way that the carrier can to solve that problem.
Olivier Lafontaine:
Yeah, the AI can simulate empathy really well better than some humans actually, but for some reason it's missing that little magic that people still require.
Brian Poppe:
Yeah, it may get there and you may eventually say like, "Gosh, do we want a human doing this or do we want the AI doing this? " That'd be a fun choice if AI gets to that level of empathetic communication as to whether or not you would want to ultimately outsource things that is probably at that customer's, one of the worst moments in that customer's life to an AI agent or if you'd want to continue to have a human on the back end of that.
Olivier Lafontaine:
Yeah. And I mean claims is an obvious one, but there's also, I think there's other situations probably that will require or that a human will be a better experience for the customer. If your payments have bounced, your payments have not passed because for whatever reason you're having financial problems, maybe that's one that's better handled slightly differently. People are embarrassed about it and things like that. So we'll se. But it's great to see that there is some... I mean, have you seen that this generates some momentum and some more excitement I will say around AI than I would imagine there is sometimes some concerns over AI, that it will replace jobs and things like that? How is it at the Mutual of Omaha? Do you feel like people are enthusiastic, cautiously enthusiastic? They're resisting a little bit? How's the feeling in general about those tools?
Brian Poppe:
Like anything, any big change like this, I think you've probably got some sort of distribution curve of folks who are really excited and all for it because they've experimented with the technology and can see how it might help them. And folks who are a bit more skeptical, perhaps by nature, perhaps by they've seen enough tech shifts that they're like, "Gosh, this is yet another one and I don't want to have to try and adopt it that way." I think in general, I guess we've got a couple of ways that we're rolling out AI and I'll give you the difference between them. One of them is the horizontal approach. So that's like everybody gets some sort of AI tool. So we're a Microsoft shop so we have Microsoft 365 Copilot AI tool that everybody has access to and can use. If you're a developer, we have developer tools to be able to help you complete coding faster.
And in general, I think the more folks actually use the tool, so one, you got to get over the hurdle of try the tool. But the more folks use it, the more excited we've seen them become about how AI can help them do their job. And then in addition to the horizontal AI implementation, as I mentioned with Copilot and others, we've got the vertical ones of which the call summarization and that virtual agent that I had mentioned are more of a vertical use case of like, well, this is going to be kind of driven by either a business unit or whoever, some business head, and here's the thing that we're going to work on with AI. And that requires some additional focus. And actually one of the things that we found to be particularly helpful is to think about that process before you just add AI to a step in that process.
It may be that, gosh, if we could get the data in this particular fashion, AI can take even more of that task than just assisting the human in doing one step in it. And so thinking about that process truly from end to end and then saying, "Well, gosh, once I've cut all those pieces out, where can AI apply and what chunk can they take?" That's been one of the more interesting things that we've learned as we have taken on those vertical use cases thinking about process first and then AI second.
Olivier Lafontaine:
No, that's interesting. Yeah, I've heard that a couple times. It's not necessarily a good idea apparently to throw AI at a bad process because then that becomes you're automating something that maybe shouldn't exist at all. And that's going to cost you still because the AI isn't free. You still have to pay for tokens. You still have to pay for whatever mechanism the vendors are charging you for that and you're going to be dependent at some point on those AI agents. And I guess that's another question perhaps I would ask. Have you started to see or some concerns over vendor management or the cost of AI as you deploy those things? Is that yet a challenge that you guys are looking at or at the moment we're still in experimentation phase and so we'll see about the cost later type aproach?
Brian Poppe:
We are probably somewhere in between. So vendor management, if you're bringing in a vendor at a known cost, known annual fee or whatever, those typically get the regular amount of scrutiny. One thing that we have though about as you are maybe using tokens in AI, we are probably, and I think actually most carriers, I don't think this is unique to Mutual of Omaha, are almost too punitive on, well, that's too expensive just based on some sort of unit economics that you're trying to think of ahead of time. And I actually think that we're too punitive and I think this is not just a Mutual of Omaha issue. I think this is an insurance wide thing of until you get a call from your CIO saying like, "Oh man, you've completely blown the IT budget because you've spent X million dollars on AI tokens, you probably have not experimented enough with AI." And we haven't arrived there yet.
We've said, "Well gosh, this seems expensive." And it's like, I don't know that I know that yet because I don't know how much of this particular task, if you're doing that, the vertical AI use cases, as I mentioned, the AI can take. And it may be worthwhile for me to be able to say, "Gosh, I can make AI do this whole thing." And even though the tokens seem expensive upfront, if it can take on a full end-to-end process, it might be worth it to us. And so I guess I would caution that anybody who works at either my company or elsewhere, don't kill that too early. The technology is probably at an infancy stage that, in my opinion, cost optimization, you don't want to do that too early because you'll probably rule out some capabilities that you actually want down the line.
Olivier Lafontaine:
Yeah, I agree. We're probably still at a stage where we should push people to use it more. When we do get the call by the CIO to say, "Hey, this is getting out of hands," maybe that's the time where we reflect on cost a little bit, but I don't think many companies in the insurance space, at least in the life insurance space, I don't think many companies are there yet.
Brian Poppe:
I have not run into anybody who has gotten the call. And I've actually been on stage a couple times and asked folks like, "All right, who's killed something because of cost?" And then the second question is, who had their CIO call them because they blew the IT budget and nobody's like, "You know what? We killed it before we even got anywhere close to that. " We are in the five-figure range or six-figure range for carriers that... I mean, if you're an individual, that's obviously way too much money to be spending on AI. But if you're in a carrier phase, you're like, "Well, that doesn't touch much of our IT budget rather than the seven or eight figure range." And if you get up there and your CIO's giving you call, then you probably actually want to listen and you're like, "Well, how do I optimize this cost rather than just send it?
Olivier Lafontaine:
" And we'll probably hear more about this as we get more into the autonomous agent type stuff. At Equisoft as a software company, we have some developers that use coding agents and a couple people can ramp up the tokens really fast. I think in your own, every insurance company has some software developers for something or other. Maybe eventually I think that's the first part that will show up on the radar. But eventually if we use the agents for the insurance operations as well, I think that's more perhaps where we'll start to see these things go in the loop and it racks up the tokens pretty quickly. And as the models become more powerful, they're more expensive, we'll see. But I don't think we're there yet. So it's interesting.
Brian Poppe:
I agree. I agree with your stance there. I did hear an anecdote the other day that there are engineers who are claiming token usage as a point of pride of like, yeah, I use this many tokens. And obviously that's easy to do. I mean, I can make Claude, ChatGPT, whatever, go generate a bunch of videos and just burn through a bunch of tokens really, really quickly. What you actually want though is tokens to productivity. And I don't know how to measure that yet because it's not even like lines of code. I thought that was also probably a pretty bad measure of productivity, mainly because a well-written program probably has fewer lines of code than a poorly written one. And yet if you're going to say lines of code is the measure of productivity, you would value the poorly written one, which seems crazy to think about.
So I don't know how to measure that yet. It's like tokens per value and the value being in our case, probably business value on an insurance basis. There's just not a good way to measure that yet.
Olivier Lafontaine:
I was at a conference last week actually, and they talked just about that where the measures are typically token consumption and lines of code and those don't really represent outcome value or value of outcomes. But then value of outcome is a nice thing to talk about, but it's really hard to measure. You can ultimately calculate dollars, but even that, I suppose, as a life insurance company, how would you measure the value of your call summarization? For example, I'm sure you can do a spreadsheet, but have you done anything to measure the value of that investment and project or it has to stay anecdotal?
Brian Poppe:
That one was a little easier to do because everything's sort of tracked in a call center. So you can say, "Well, I've saved this many off-call minutes and I can clearly see it from the work that they've done." And so then it's like you can eventually translate that to an employee salary and compare that to what you spent on AI. Someone has an easier ROI, but to the point that you're making, we looked at the outcomes, not necessarily the lines of code or the token usage it took to ultimately get there. You're saying, "Well, I can actually isolate the particular AWS account." We have AWS as our partner. We've been public about that. The particular AWS account that is racking up the call summarization charges and compare that to the productivity measures that you measure typically in a call center. And that's the ROI that we've got there.
It does get a little squishier though because even if you write... So imagine you're building a, I don't know, some sort of web portal or an application or something and you're like, "Man, I really optimized the token usage. Everything worked out great. The experience is great." And you still aren't getting the customer throughput that you want. Well, is that a webpage problem? Is that a marketing problem? So there are some other confounding variables that even if you're trying to be diligent about measuring dollars of value to token usage, still it's hard to isolate in that case.
Olivier Lafontaine:
Maybe the lesson learned is that if you have an easier way, if it's a business process that has an easier way to measure outcomes, then maybe those are the better scenarios to automate with AI in the first place. So we'll see how that evolves. Like I said,
Brian Poppe:
That's one that I was wrong about. I'll freely admit that one. It ended up being a great use case for exactly that reason of relatively easy to measure the value and payback relatively quickly.
Olivier Lafontaine:
Well, that was very interesting. Maybe before closing, if you had some advice to you, you gave a couple points, but any final advice for our listeners around modernization and running technology changes either in the direct to consumer space or other life insurance lines of business, what would be your best advice considering your latest projects?
Brian Poppe:
A couple pieces of advice I would give is if you haven't toyed around enough with AI, would highly recommend doing that. At the moment, Claude seems to be the best of the major LLM providers. Would recommend trying Claude Coke and seeing what an agentic AI can do for you and probably starting in your personal life. I would not recommend that you go around any IT controls and try and put that in your work environment. So that's probably one. I'll actually use the technology and then you'll start to unlock some other ideas that probably do translate over to work and you can go talk to your chief information security officer and probably your CIO to get that implemented. Second thing I guess I would consider is you do actually have to start with process. Olivier, to your point, you really don't want to apply AI to a bad process.
If you can just avoid that process altogether, that's probably actually the best outcome rather than optimizing an AI's token usage to execute a bad process. So even if you're like, "Gosh, I deeply know how to use this technology and I can apply it here," you're going to learn a whole lot of things if you start to think about that process end to end. Those are probably the two best pieces of advice I could give, regardless of whether you're applying that on maybe a software migration or if you're thinking about applying that in some sort of product development or anywhere else that you might consider call summarization, for example.
Olivier Lafontaine:
Some operational process or improvement in cost structures and things like that, right?
Brian Poppe:
Yeah. Any of those would value looking at the process first before you start just applying AI to a step in it.
Olivier Lafontaine:
That sounds very logical and good advice. I would second that every day. Thank you for your time. This was a very interesting conversation. Hopefully we'll have you on the show in the future again in a couple years perhaps as you progress through those projects and have new thoughts around what to do in the industry. And maybe I'll see you at one of the InsureTech conferences or hear you maybe a DJ for the event or an after event.
Brian Poppe:
Yeah, that'd be great. If you find me at one of those, be sure to interrupt me and say hello. I'll look like I'm busy, but I'll for sure take some time to say hello. Thank you again for having me. I had a great conversation today and I look forward to coming back soon.
Olivier Lafontaine:
All right, thank you. Ryan's perspective is a reminder that successful transformation starts with solving the right problems. What stood out to me most was his focus on process first and technology second.
Whether discussing underwriting, customer service or AI, the goal isn't simply to automate work, it's to create better outcomes for customers and employees alike. As new technologies continue to shape our industry, the disciplined approach may prove just as important as the technology itself.
Thank you so much for listening.
