"You Can't Just Slap AI On It" — Brian Poppe's Four-Phase Test for Carrier AI Adoption

Paul Tyler (00:02) Hi, this is Paul Tyler and welcome to another episode of the L&A Hub. We've got a great guest today — one I bet a lot of you know. Brian, tell people who you are and what you do.

Brian Poppe (00:17) Hey, Paul. Brian Poppe. I work at Mutual of Omaha — and that's about as cleanly as I can say what I do on a regular basis, because if you ask me in a year or two, sounds like I'm probably going to get another job. I've been at Mutual of Omaha since 2009, and as my intro suggests, I've done a variety of things in my time here. I started out as an actuary — actually at Lincoln Financial Group for a couple of years — and moved over to Mutual of Omaha in 2009. Started doing long-term care pricing and product development. Worked in risk management for a period of time. Worked in innovation, which is probably how a lot of folks who recognize me have seen me — at some sort of conference related to InsurTech. Managed a P&L for a couple of years. Moved into technology. Ended up as Chief Data Officer. The tenure of a CDO has got to be like a year or less, and mine certainly fit that bill — because the business came calling again, and now I'm back managing a P&L. I don't have all the badges yet at Mutual of Omaha, but I've got to be getting close. I think I haven't had a stop in finance yet, but most other places I've had a short stint.

Paul Tyler (01:33) Well, you've also had a stop on the dance floor. You're also a talented DJ.

Brian Poppe (01:38) I am, yeah. That's a long-standing hobby. One of my high school friends and I started a business in high school and I just sort of never stopped. So if you're at an InsurTech event, you may see me at one of the after-hours parties on the decks.

Paul Tyler (01:55) That's a skill I would love to pick up, but I'm not sure the people on the dance floor would want me.

Brian Poppe (02:03) It's definitely gotten easier as technology has gotten better. Although I've gone all the way around — I recently picked up vinyl DJing, because I think that's a fun visual effect for the folks watching.

Paul Tyler (02:14) You mentioned innovation. Ideas are cheap. In fact, even great ideas can be cheap. So when you were running the innovation practice for Mutual of Omaha, what happened to the ideas that didn't make it through?

Brian Poppe (02:48) We collected a lot of ideas. I'm saying "we" — it was literally just me. I was the entire innovation department. I had a travel budget and a network internally that could actually help get stuff done. So I did spend some political capital trying to get some of those ideas built into the process.

There are some that we tried that just didn't work. I still love this one — we partnered with a company called In-Memory. They started in France. At the time of a loved one passing away, they would partner with the funeral home site, collect the memories that people were posting on the eulogy page, turn them into a story, collect uploaded photos, and ultimately turn that into an actual memory book they would send to the family.

I still love this idea. The challenge: when we offered this as an insurance carrier, free of charge to customers, the timing was just not right. When a beneficiary is calling in saying my mom or my dad recently passed away, they don't have the mental capacity to take on one more task. We had taken on a lot of the work between us and In-Memory to set that up — but it was still one more thing. So we learned that when you extend the offer really matters a lot, even if it is a great idea. We piloted it for about six months, tried to get adoption a bunch of different ways, and ended up killing it.

There are others where it's: we love this idea, but the lift is going to be too difficult to ultimately execute. Knowing what we know from our current tech stack, it's going to take forever — and we've got other things we want to do in the meantime. So that one ends up on the backlog until we fix the tech stack. Plenty of things end up on that floor as well. And then there are the ones you ultimately end up scaling — we piloted it for six months and said, you know what, we want to make this bigger.

Paul Tyler (05:51) Until you do this, you don't realize that to get a business working, you've got to get 20 things right. Thing number 18 doesn't quite work, you're done.

Brian Poppe (06:07) Anybody who's worked in a startup has those stories — we think we have product market fit, then we tweak this one thing, and it falls apart.

Paul Tyler (06:20) One startup I brought into a prior carrier was a great idea: what if you don't have to call HR for benefits info? This was just pre-ChatGPT, when machine learning was coming along. HR loved it. The problem: you go in and ask, well, does it do this? No. Does it do that? No. I talked to some early founders at Mint.com once. I thought Mint was great. They told me the biggest problem was: yeah, they got the data connections, but most people using it had a negative net worth and they didn't want to open the app.

Brian Poppe (07:06) If the news is bad, you just try to avoid it.

Paul Tyler (07:14) A couple episodes back I had Nick Gerhart on the show.

Brian Poppe (07:23) Nick and I met when we were members of the Global Insurance Accelerator. So I crossed paths with him a decent amount going over to Iowa.

Paul Tyler (07:34) He talked about a phrase he loves — the InsurTech Petting Zoo. Looking across the business: when does an innovation shop have a petting zoo that's theater, and when is it real? What's the litmus test?

Brian Poppe (08:01) From the carriers' perspective, we certainly started with the petting zoo. It's almost a requirement if you're working at a large carrier and trying to set up an innovation practice — you have to actually show folks what might be possible before you can get them to ever believe a 20-person company can scale to serve all of your customers.

So there is something to the petting zoo, but you ultimately want to get out of the zoo and into something else. We're not great at it yet, to be clear, but we have some wins under our belt. The pattern: as a carrier, set up the pilot in a way where it's a defined timeframe and a defined thing you're trying to learn. It might take three months to get to the phase where you're ready to actually go and ready to learn that thing. Then at the end of three or six months — however long you've deemed that pilot worthwhile — you make a decision. Sometimes the decision is, we just can't make this work. In other cases, this really ended up working out and we scale it from there.

I took my CTO out to the Silicon Valley petting zoo. After about nine months or a year of work, I'd lined up a week where I thought I could set up a full cloud insurer end to end. So I brought the CTO with me — just me and Tim. We visited a whole variety of companies. I tried to get it all the way from application to claim. I couldn't quite do it in chronological order, but I got all the pieces in place. We spent four days. At the end of that week, I said, Tim, I don't have anywhere to go with this. It's just me. I could spend the next three months trying to put these pieces together.

Tim, thankfully, was like — all right, let's invite the petting zoo to our tiny little farm. So we invited them in. We set aside three developers to run a proof of concept. Coming out of that 60-day proof, they said this is actually going to work. They didn't say "Brian, you're a genius" — but I really wish they did. Then we figured out, now how do we start putting real products that scale on this thing? That ended up becoming the backbone for our future-state tech platform. But it started with a petting zoo tour. I don't know how to avoid that as step one in the overall innovation process.

Key indications when it's real: we set aside a clear timeframe with a clear outcome we were looking for. When we got to the end, we said, okay, what's the next step? If you're on the startup side, you might be saying, what are you going to do with this pilot? What's the outcome you're hoping for? Make sure you're aligned on what success looks like. The further you can build out that road map, the more indication you've got somebody on the other side who knows what they're doing.

Paul Tyler (11:55) Let me pivot to AI. I know this is a passionate area for you. Let me check my research — you've talked about four phases of AI adoption. Phase one: awareness. Phase two: helping AI to help individuals do their job. Phase three: redesigning processes. Phase four: AI-driven processes connecting to other AI-driven processes. That right?

Brian Poppe (12:39) Good work. You ready to go on stage? That's been a couple of my talking points lately. But yes.

Paul Tyler (12:48) It feels like we're going through the same path here. Nobody understood what AI was. More and more, they do. But still, I walk into these meetings — even at technology companies — and the knowledge level is so jagged. One person knows nothing. Somebody else is going to talk to you about recursive training.

Brian Poppe (13:19) I was just in a meeting yesterday where a couple of team members were like, "yeah, I run my own open-source server. I forked the code and wrote my own. I run that on my machine at home." And somebody else was like, "yeah, I just use the base version of Gemini, I don't pay for anything." To the jagged level — there's somebody who's super deep into it, and somebody else who at least is at level one, maybe thinking about level two.

Paul Tyler (13:49) So how should carriers approach working through these phases of AI adoption? By the way, phase four — we can talk about it. I'm not sure what phase four is going to actually do.

Brian Poppe (14:08) You and me both. There's probably a few ideas of what phase four might look like, but I don't know that there's anything in insurance or financial services that looks like phase four yet.

You have to start at phase one. I don't know that you can skip the awareness phase. It's: I need you to try this. Don't worry about making it perfect. Don't worry about being able to one-shot prompt or whatever outcome you want — just make it available. Provide some base level of training of "here are some things you might want to try."

We're a Microsoft shop at Mutual of Omaha, so everybody has access to Copilot Chat. About 70% of folks have access to Copilot — which is the second version. Microsoft's naming is atrocious, by the way. Then you've got Copilot Agent above that, and a few folks have access to that because they're trying to do level-three things — building processes around it. So you've got to start with level one, the awareness.

Once somebody has a little understanding, a good prompting from a manager is, "have you tried putting your bullet points into Gemini, Copilot, or whatever, to create that email or that report?" And actually making it okay as a manager to say, I know you're going to use AI. I've talked to college professors recently — I was asking professors and students how they're using AI. If they're ignoring it completely, that's probably a bad sign, because the students know about it and are using it. Both the professors and students were like, "no — I tell them, go use Cursor to do this coding assignment." So at least there's some input there. That's a good parallel for a manager: if I ignore it altogether, I'm going to end up with employees out there using it anyway.

Stage two has got to be: if you're in a leadership position, make it okay to experiment. If you don't like the output, you might call them out — "you missed this thing." I had a situation where one of the Copilot agents made a presentation, and I asked a subsequent question later, and it referenced its original presentation. And I was like, hold on, you're not quite hallucinating, but you don't get to double-count the work that you did. Give me a new reference.

Once you get to process redesign — that requires something similar to what many of you have probably gone through with process transformation. There's no real difference in the way we've thought about process transformation before AI and after AI. It's just that now I've got a new tool that allows me to automate more steps easily, or that can help bridge the gap between the deep human work that might have happened and the actual process they want to execute.

You can't do everything at once in phase three. And to your point, phase four is a complete unknown. Phase four is trying to really connect AI to another AI to execute a process. In the insurance context — I've got a producer AI out there trying to help a customer, and they're trying to buy from my wholesaler AI at the carrier — connecting those two pieces together to get the customer the thing they want, easily, quickly, without really any input. Yet to be determined.

That's a good thought experiment. You can already book — OpenTable is maybe the best example. In ChatGPT you can tell ChatGPT to go book you at a restaurant. OpenTable has an MCP server that the Copilot agent can connect to, and then it just books directly. You as the human are saying, "just book me a table." Now imagine that world, but in banking or insurance, and you've got a real interesting problem to solve where you've got AI talking to AI.

Paul Tyler (19:29) Even just phase two to phase three is a huge leap. Put a tool on a desktop and — at least the version of Copilot I had was horrible. "Hey, can you schedule a meeting for me?" "Yes, I scheduled a meeting." No, you didn't. "Oh, whoops, I didn't."

Brian Poppe (19:52) There's a big gap there. Maybe we add a 2.5 to that four-phase framework.

Paul Tyler (19:58) I'm on the LIMRA AI governance group that's been formed. There are a lot of companies on it. Some of the big carriers are way behind. Some of the small carriers are way ahead. I haven't yet figured out the common denominator. So far, what I've seen is a lot of work in phase two — some more sophisticated than others. "Hey, we put enterprise ChatGPT or Copilot in. We've got usage." Or, "we've put a tool that's geared specifically to a certain workflow — probably a multi-step prompt, maybe querying data to write letters, write correspondence." I haven't yet heard a lot on the process redesign side. Have you? What area do you think will be the first significant process redesign as a result of AI?

Brian Poppe (21:38) There is something to back-office service. There's probably something there that's less risky. An example might be NIGOs — not-in-good-order applications. As applications come in, that feels like one — somewhere between phase two and phase three. Could I have a smart agent that can solve some of that problem? Maybe. If we restructured some of the NIGO process and built it with AI at the core, that would make the agentic AI's life much easier, and probably process it faster than using whatever system we've got. That feels like a good one — and it would be safe regulatorily. That comes up a lot — how is a regulator going to view this? What happens if the AI gets something wrong? Nobody's happy in that situation.

Paul Tyler (22:47) I've spent a lot of time with underwriters lately. You'd think on the surface, underwriting would be the first area to make a massive process leap. But to your point, it's also the most highly regulated and the area with the greatest amount of risk. You're not necessarily talking about applications per hour — you're talking about embedded value of a product over a significant amount of time.

Brian Poppe (23:20) You've got that, and underwriters by nature are some of the most skeptical folks in the company. They're a tough group to get on board.

Paul Tyler (23:26) They're really, really tough. Now fast forward — you're actually running a business. What's it like to go from solving the problems to solving a P&L?

Brian Poppe (23:45) I manage the individually sold life insurance products at Mutual of Omaha. There are a lot of new sorts of problems that crop up. A lot of the things I learned as Chief Data Officer about infrastructure, about how to think about analysis or even data science, are helpful in solving some of the problems in managing a P&L — whether that's sales challenges or distribution conflict.

We have all three kind of distribution channels — direct to consumer, our Mutual Advisors, and an independent distribution network. Sometimes those folks cross paths and you create a distribution conflict. That requires some external skills. It's not just "I've got to apply technology to solve this problem." Some of it requires finesse — there could be routing you do, commissions paying differently, whatever. There's a variety of ways to solve that.

I've seen the technology from the inside in my time in IT. So now you're trying to prioritize big initiatives to solve the main bottlenecks you see. Between the external pressures that come with a P&L, the internal pressures from finance, and the no shortage of technology challenges or data challenges — it's a fun problem-solving exercise. Probably one of the best things is being able to directly impact customer experience, and have a team aligned in figuring out how do we solve those customer problems as quickly as we can.

Paul Tyler (26:00) Let me poke on that. You've talked a lot about simplifying products. Zinnia's tagline is simplifying insurance — principle is great. But we're in an industry that's grown on differentiation. How do you square that — running a P&L, you've got to have a product that stands out, but from your background, you know the importance of standardization and why simple may be better in the long run?

Brian Poppe (26:47) Each company is going to answer this question differently. Think about the strengths you have and the situation you're in. For Mutual of Omaha — we're a big player in simplified issue. The way you win in simplified issue is a very easy-to-sell, very easy-to-buy process, which lends itself to your point about a simpler product. You can win on convenience. If you're a heavy, fully underwritten carrier, you'd say, that's not it — it's about how do I provide the most value to the customer on a cash value basis.

You have to think about: what's the thing we do well — and lovely if it's better than anybody else, but at least what's the thing we do very well. We do simplified issue well. We don't really want to give that up.

When we get to the fully underwritten side — because we offer those, too — one of our differentiators is that we're a mutual company with portfolio cap rates. So we have very stable cap rates compared to some folks who use new money. The fact that we're mutual gives us a longer runway. You'll hear us talk about: we're here for the long haul, and we can't promise the illustrations, but we're going to deliver pretty close to them. We don't play games with illustrations trying to be number one on the chart. We're trying to deliver the best outcome for the customer over the life of their policy, rather than how it appears at day zero.

Those two models are probably the things we spend the most time thinking about, then building products — whether insurance or tech — to support those two main paradigms. To your question about squaring with standardization — it does drive the technology underneath. You're trying to position the company in a way that builds on those strengths. That helps you make tech choices down the line. We've made some trade-offs we did weigh against those as the strategy.

Paul Tyler (30:05) As the world's gotten more digital, the definition of product has broadened to include not just what shows up in the illustrations, but the experience. How easy is it to sell? How easy is it to actually service after you've sold?

Brian Poppe (30:22) Talk to simplified issue producers — absolutely. The product is the whole thing, from how easy is it to quote to how easy is it to hit the apply button with a customer there. And to your point, how easy is it to service down the line. That's going to matter a lot, particularly as you get to that phase four.

Paul Tyler (30:43) Last question: clearly the industry is going through a huge transformation. How much of that, as we get to phase three or phase four with AI, is going to be a technological transformation, and how much is going to be an organizational change?

Brian Poppe (31:07) Very good question. In our experience, it's a lot of both. I don't know which one's heavier. You're going to get some upheaval, almost, on the org side — whether that's internal or society at large as AI becomes more prevalent. I wish there was a cleaner way to do it. The best parallel is probably the adoption of the internet. We probably got rid of some things that did not need to exist with the internet — pay phones, and probably all the people who worked in the phone industry. But we also created a whole bunch of new jobs that did not exist either. AI may afford us the same opportunity. I don't have a better parallel.

That's pure org change. The tech adoption is maybe a little easier. As I noted, you can't just slap AI on anything, but you can start to think about processes and designing them with AI at the core. That requires some technology change, but a whole lot of org change.

Paul Tyler (32:31) When I started my career as a management consultant, my very first project was for AT&T on their operator services business. Dial zero. Late '90s. That was a $2 billion-a-year business.

Brian Poppe (32:48) Wow. Now they've probably got just one operator to run everything.

Paul Tyler (33:00) The biggest market — the most lucrative — was prisons, because you had to dial an operator to dial collect. When was the last time you talked to an operator? Never.

Brian Poppe (33:17) Never. I have at some point in my life, but it's been a long time. I don't even know if you could get one on the phone at this point.

Paul Tyler (33:25) I should try dialing zero, see who answers. Hey, listen — this has been great. We've been involved in the Founders Chair, we see each other at ITC and Omaha events. How do you stay in touch with trends? If you were giving advice to people at carriers or distributors — how do you stay on top of this? It seems to change every single week.

Brian Poppe (34:00) Depends on the type of trends and how far you want to go. Personally, I have a couple of AI-forward podcasts. They're short. I don't listen to all of them — I don't have time. I may send you some names; you could post them in the show notes.

The second: I have a network of folks I draw on. At a conference like ITC, I typically reach out ahead of time and just say, hey, if you're going to be here, I'd love to catch up for 20 or 30 minutes. Then sit down and chat about: what are the things I've seen, what are the things you've seen? Together we can both learn something and come back and experiment more.

One of the ways I came up with this four-tier deal was out of a conversation with a friend who works in venture capital. That's been a helpful framework — articulating both externally and internally where we are on the adoption curve. If you want to get here, here's what that looks like. Then you start talking about process transformation as one of the things.

You'll eventually find your folks. I find them through conferences — that's the easiest. Some other folks you could find through online forums; Reddit is probably a good source for folks who might have a similar interest. Those are my two primary ways — the podcasts and the network of folks I've met over the years, because I like the way they think.

Paul Tyler (35:46) Brian — thanks so much for making time on a late Friday afternoon. If people want to reach out, what's the best way?

Brian Poppe (35:56) Easiest way is to find me on LinkedIn — linkedin.com/in/bpoppe. Otherwise, just search Brian Poppe. I'm sure I'll pop up.

Paul Tyler (36:03) By the way, you have a podcast too, right?

Brian Poppe (36:14) I host a couple. There's one I host from work called Tech Talk, where we highlight folks in the IT department who are working on cool things or have had really interesting career paths. We do both. Those publish about once a month. Then there's a local podcast I host with Shauna Dorsey, where we feature entrepreneurs and folks in technology around the Nebraska area. I usually post both on LinkedIn. Follow me or connect with me, and you'll see them.

Paul Tyler (36:49) Thanks so much.

Brian Poppe (36:51) Appreciate the time today, Paul. This has been fun.

---

Thanks for listening to the L&A Hub. If you enjoyed the episode, follow the show wherever you listen and leave a quick review to help others find it. Until next time, keep testing, keep learning. Brought to you by Zinnia.