When AI Learns Your Workflow Who Wins The Market

Your AI demo is easy. Shipping AI inside a regulated insurance workflow is the hard part and that’s where the real advantage gets built. We sit down with Doug Marquis, CTO at ZyWave, to get specific about what’s changing in insurance technology and why “agentic AI” only matters when it reliably drives growth outcomes for agencies and brokers.

We dig into ZyWave’s front office focus from lead identification and marketing to quoting, service, and renewals and why that matters as the insurance workforce ages and capacity tightens. Doug breaks down the two inputs that separate winners from copycats: access to high-quality insurance data and the discipline to encode domain knowledge and real producer workflows into artifacts that AI agents can actually use. We also talk through how automation can cut the administrative load so producers spend more time being consultative with customers.

Then we get practical about enterprise AI: compliance and security up front, licensing and data access controls, and the “AI harness” you need around agents, including observability, testing, cost management, and explainability. Doug also explains ZyWave’s insurance MCP server and how plugging insurance-specific data into LLM tools turns generic answers into specialized, contextual guidance. If you’re a CIO, CTO, or product leader trying to move from pilots to production, this is a clear look at what to prioritize next.

If this helped, subscribe, share it with a teammate building enterprise AI, and leave a review with the biggest challenge you’re facing right now.

Transcript

Welcome And What’s At Stake

SPEAKER_00 0:02

Hey everybody. Interesting chat today on how AI is changing insurance, what it takes to move beyond the demo stage, and where AI in the enterprise is headed next with ZyWave. Doug, how are you?

SPEAKER_01 0:14

I'm doing great, Evan. Thanks for having me today. Always uh exciting to talk about AI. It's been uh it's been quite a fun journey over the last couple of years here.

SPEAKER_00 0:23

Indeed. And we're just getting

ZyWave’s Role In Insurance Growth

SPEAKER_00 0:25

started. Uh for those who may not be familiar with you or ZyWave, tell us a little bit about yourself and how do you describe ZyWave these days?

SPEAKER_01 0:32

Yeah, so uh I'm our chief technology officer at ZyWave, and really that encompasses three things. It's managing our software, our SaaS platform, a whole bunch of data that feeds that and feeds a lot of the AI, uh, and also a lot of content that we develop, which you can see also feeds into some of the generative AI stuff we do. Um, my background is really in enterprise software. I started at big consulting companies and then went into venture capital and now private equity-backed software companies. And it's probably been about two decades ago where I kind of had my first foray into machine learning and natural language processing and sort of the traditional mode, you know, traditional sense of what AI is. Um, so uh it's been fun to see that all the way through the internet era, the mobile area now, and there's this kind of AI era that's out there. Um so I feel like a lot of the things I've done in the past in the enterprise are kind of all coming to fruition within this um new transformative technology that's been kind of put in front of us. So that's the exciting part, and that's some of what I've done. In terms of ZyWave itself, we're one of the largest platform players in the insurance space. And we really focus on what we call the front office. So a system of action, a system that can essentially, from a growth perspective, help insurance agents and insurance brokers uh really drive everything from their lead identification through marketing campaigns, through quoting business, ultimately servicing those customers and renewing that customer. So, how do you win more business? How do you retain more business? How do you grow your business as an agency or a broker or producer? And that's really where we play, a little bit less so, or not as much kind of in the back office space, the system of record, the account management systems, uh, underwriting, claims management, not really where we focus. So that's what we do. And to kind of give you a long answer to that question, I mean, some of the problem that we solve there is a lot of the insurance industry wants to continue to grow. The folks within those industries are kind of aging out a little bit. There's not very many people coming in, and so what you end up with is this scenario where technology and automation have to fill that gap. You want to continue to grow, but you've got fewer people coming into the industry, and that's a real core problem that we see across our entire customer base today. And that's ultimately what we're trying to help with with our AI platform that we're talking about.

Data And Domain Knowledge Advantage

SPEAKER_00 2:44

Brilliant. And you know, the AI challenge opportunity. Um, where do you think companies in this space are are able to build a real competitive advantage with AI and with ZyWave?

SPEAKER_01 2:59

Yeah, I think there's there's a real there's a couple of things that are incredibly important to that. One is uh getting at least having or getting access to a lot of really good data. Um within every industry, data is a problem. The quality of the data, the completeness of the data, the breadth of the data is a problem. And so you've got to find out where you can get that, either internal within your enterprise or through third parties and other folks. And that's a real key component to things. I think the people that are being successful have have started down that journey and figured out a little bit of that. I think the other side of the coin there, though, is uh really uh taking the specific knowledge and the specific domain and use case information and and and codifying that so agents can actually use it. Right. So understanding what a producer does on a day-to-day basis or uh a different role within your company and how they do that and codifying that so an agent can actually use it. And so building on that and continuing to evolve that as things evolve, those are really two key components to making sure that the agents can do what you want to do with the data that's necessary to ultimately drive some outcome that you're trying to achieve. And so, from a core component perspective, those are the things that we see uh we're trying to put together within our software system to allow people to then leverage that across uh across their growth initiatives within their company.

Automating The Producer Workday

SPEAKER_00 4:14

Brilliant. So in the enterprise, as you know, it's all about the workflow and in insurance, particularly what is what is the workflow that can unlock via you know an AI assistant, things it can do reliably on its own. What are you seeing at this page?

SPEAKER_01 4:30

Yeah, again, I'm gonna I'm gonna kind of focus on what we do with our ZyWave Apex platform. And again, we focus on that front office area. So if you think about what a producer does every day, 60% of what they do is largely administrative work. So they may go say, hey, look, who who are the ideal customers that I can go after? And I figure out who those customers are that they can chase. Who are the contacts within those customers that I want to send an email to? They craft an email or an email campaign that they set up. They eventually sort of go through this process of, you know, and then looking at the risk that somebody might have or the reason why they might be interested in buying. So we've built agents across that workflow to really automate a lot of what's there. So we we come up with the ideal customer profile, we we do analytics to see who they can best sell to, who they will have the most likelihood of success based on lots of information, policy information, when do you renew? Have you had any OSHA violations, all kinds of stuff like that. Uh, and then we bring that into a personalized marketing campaign with generative AI that generates the emails who are very specifically and targeted at the individual, not a blast email that's out there. Um, we see click rates that are two, three, four X, the industry average. You know, I'm built around quick rates on those types of emails because of the specificness of it and the timeliness of that. Um anyway, we take that all the way through essentially at the end of the day, running quotes on the business and putting together a package that the producer can then go sit down with and do what they're best at, which is hey, I want to be the consultative seller or the consultative um person to help you determine which coverages you should buy, what the pricing should be, uh, and kind of iron all that stuff out. And so we really that 60% of the administrative duties can't go significantly down with the agents that do the work. And that's where, again, that's where our our our sort of uh software plays and when we're really focused on trying to help the help the industry move forward.

Compliance And Observability For Production

SPEAKER_00 6:18

Brilliant. So agentic AI is all the rage. We're hearing about it every day, top top of the fold, as we used to say. Um, I'm using it in the background now, uh Claude Cowork and and Manus doing various things. Hope it doesn't crash my computer as we're chatting. Uh, but what separates my work as an individual or you know, a fancy demo from a system in the enterprise that can work reliably at scale and you know in production? Yeah, I appreciate the the above-the-fold reference.

SPEAKER_01 6:49

It kind of grounds me in the it grounds you and I in the same era of uh you know our likelihood. Uh but I think there's a number of things there when you think about how do you get to production. I mean, look, one of the things that I think is most important, especially in insurance in particular, any regulated industry, is you you really need to think about the compliance-related uh um needs of the system and the security-related needs of the system up front. And so when you have these agents out there running, for example, you really have to know who they're running on behalf of, so they're, you know, so that you don't end up with somebody accessing the wrong data or or producing something that they're not qualified to produce because they're not licensed for it, or they're not able to sell in a certain state. Variety of things that come into play around this. And so I think when you think about you know the difference between kind of that demo environment versus a production environment, it's easy to get away with kind of uh, you know, medium, medium quality data. It's easy to get away with maybe taking some shortcuts on the compliance front because there's not that much risk in the security front. But I think as you get into production, you need to have a lot of that. And so that's really important. I think the other thing that's probably important uh just at a more macro level is kind of what we call the AI harness or the thing that kind of surrounds all these agents working. So you have things like observability, and you can explain to a regulator or a customer or your internal testing team how did we get to that result? How did we get there over time? And that's a little bit easier said than done in this environment where we're using third-party LLMs for the most part, and those LLMs can change every day. So the thing you ran today and got an answer to, you may run that tomorrow the next day, and you don't necessarily get the same answer for a variety of reasons. Um, one, because it's a bad determinic system, two, because they could have changed something within their system in terms of how it learned. So understanding how you got to an answer uh in this sort of world is is really important from a number of perspectives. And that actually for us ripples across our whole company in the sense that customer calls in to support, support needs to be able to answer that question. Compliance officer comes in, we need to be able to answer that question. Right. So and then just from a testing perspective, to make sure the quality is there, we need to answer that question, and we need to monitor it over time to make sure we can always answer that question and get those sort of reliable results. So, you know, those are a few of the things I think that fit into the more um enterprise-grade AI systems versus the demo grade uh AI systems that you'll you'll you'll play around with.

What An Insurance MCP Server Does

SPEAKER_00 9:11

Interesting. And you also recently launched what you call the industry's first MCP server for insurance. Um, what does that mean exactly in a world of many MCP servers out there available? Why does insurance need a specific uh MCP?

SPEAKER_01 9:28

Yeah, that's uh that's a great, great question, Evan. And um so we have, as I mentioned before, a lot of data that that under that underpins our ball system. And so what this uh MCP servers essentially allow you to do is instead of going into a generic um club code or quad co-work or open open AI, ChatGPT, or any of these sort of tools that you use to just type in a prompt, normally what you get back is a generic answer that's trained on internet data. And look, some of those models might be a little better than the other, and they leapfrog each other over time, but generally speaking, they're kind of a commodity. Everybody has what everybody can get kind of that same answer back. And so when they're NCP tools, what it allows you to do is plug in all of this insurance-specific data into the LLM. So if you're, for example, um talking through Claude or through ChatGPT, you can ask it a question and it has a whole foundation of insurance knowledge that it can come back with an answer for. So it goes from sort of generic answers that are trained on the internet to super specialized, you know, insurance-specific knowledge about an individual, about the industry, about what's happening out there. So these MCP servers essentially kind of supercharge it and really get down to the specifics of um things that might be specific by states, things that might be specific by user, but certainly things that are specific from an insurance perspective. So putting all that information into the LLM just kind of raises the bar in terms of the intelligence that the LLM has and the quality of the answer that you get back.

SPEAKER_00 10:56

Wow, I bet. So uh fascinating

Customer Results And New Use Cases

SPEAKER_00 10:59

stuff. Can you share any stories or anecdotes or insights uh into your customers using AI today, maybe examples where there are some real tangible results?

SPEAKER_01 11:10

Yeah, I mentioned uh we continue to monitor our Apex launch and what we call our producer agent, our advisor agent that are on the market uh as of this last week. Um for the second one of those. The first one got released uh earlier this year. Um but we see again across the marketing landscape, again, open rates click rates that are of the mortar magnitude I mentioned earlier, three to five X higher. Um we have um, I mean, in specific details, we have a number of customers out there who have started to use our MCP servers, and they're using those in use cases that we hadn't really even imagined. Uh, but there's some really interesting stuff that they do within the workflows within their companies that allow them to really take this data and kind of plug and play what they want to plug in play. So we we accomplish or we can match that whole workflow that I talked about. But in many cases, some of our customers they may want to have us identify the customer profile, but they want to produce the briefing that they sit down with the customer on. So they can pick and choose the different pieces that we built and plug in the different pieces into that, integrate their own MCP servers on for this into the into the mix for additional data sources and make things smarter and then pull all that stuff together. So in our larger enterprises, we see a lot of that. Um, I guess um the sort of uh use cases that we probably didn't think about and aren't necessarily directly supporting our product, but you can certainly build them on top of those types of technologies. And that's really why we did it was to allow that sort of flexibility out there in the market for folks to do a variety

Build Versus Buy For Agentic AI

SPEAKER_01 12:36

of different things. So, you know, at a high level, um one of the things we've heard around, hey, what should we be doing with this or how should we be using from a use case perspective? I think one of the things that's important to think about is what do you want to do as a business? So if I'm a CIO or a CTO out there, uh a lot of folks went through the the path that you talked about, which is hey, why don't we do some bills? Why don't we get some prototypes and proof of concepts put up? And then they smartly and intelligently took a step back and said, okay, is this the business we want to be in is owning and operating these types of identity systems going forward? Or, you know, do we want to draw the line and say, look, we want to be very specialized and specific to insurance, and we'll look at our technology vendors and our technology people to kind of support some of this stuff. Because as you mentioned, there is a big sort of delta and leap between uh those two areas. And I think a lot of people are making different choices, and there's no wrong choice with that. It's just a question of what's right for you as a business. Do you want to focus on pure insurance or do you want to get a little bit deeper into the tech or a lot deeper into the tech, uh, depending on where you think your strategic priorities are? So um a little bit of a tangent from the question you asked there, but uh uh I think it's important to kind of understand that a lot of people are doing things differently and taking different strategies to the overall approach to how they use a Genetic and whether or not they build it or buy it.

SPEAKER_00 13:53

Brilliant.

Practical Steps From Pilot To Production

SPEAKER_00 13:54

And you know, on that note, any advice to technology leaders on getting from AI pilots into production deployments, either within insurance or outside other sectors, having gotten your hands dirty and I'm sure burned a few times at least. Uh, what what's your wisdom to them?

SPEAKER_01 14:12

Yeah, and it, you know, and um um it takes an iteration or two to get things right, right? So again, we started a couple of years ago on this, and we didn't necessarily get it exactly right to begin with. We didn't necessarily have all the right skills in-house that we needed to move at the pace we wanted to move. And so, you know, we've addressed a lot of those issues at this point. But um, you know, to your point, there's a there's a number of things out there, and some of this I think people are gonna be like, yes, I've heard that before. But but it is true. I mean, part of it is starting with understanding your business outcome and what you want to have the business outcome be and be able to measure that uh because you've got to stay focused on stuff as technology can take you a lot of different directions, and there's a lot of shiny objects and a lot you can do with that, but really you've got to understand what you're trying to achieve as a business as the first step. And then secondly, I think the thing that is probably the limiting factor in most cases is a lot of a lot of the, like again, like I said, having the data in the right place that you can use it. There's a very big difference, for example, between an individual running a certain database transaction and an agent running that same transaction 30,000 times a second, just from a pure technology perspective. The scalability, the constraints, and the capabilities of the systems that you built may not hold up to what you had in the past. So understanding where your data is and being able to get to that data in an agentic way is really important to be able to allow sort of the product teams and the people who are building capabilities on top of this from an agency perspective. Um, it's important to have that in the right place. And then the third thing I mentioned to you before, which is you just got to be thoughtful about the harness or the thing that floats around this thing. So you understand the things that I mentioned, like observability, uh, you understand your costs, you know, and doing things in the cost effective way, you understand whether you're getting the metrics that you need to have, your security layers in place. You know, when you think about the things that are there, there might be a few more things than you've had in the past that need to be supercharged, but when you think about traditional SaaS software, you got to think about security. You got to think about logging and observability, you've got to think about um how you're gonna manage compliance within there. So um, I think some of those things have bubbled up a little bit and they're slightly different in terms of how you think about them. Um, but all those things in traditional technology are important here, and then you know, packs too in in some cases, I guess.

SPEAKER_00 16:25

Yeah, well, that's what's learned. Uh fascinating.

Roadmap And Where Enterprise AI Goes

SPEAKER_00 16:28

Looking ahead, can you give us a peek into your roadmap, perhaps, or you know, where do you think the industry's headed, either within insurance or maybe the bigger, broader software industry over the next one, two, three years?

SPEAKER_01 16:40

Yeah, uh we have um a number of new agents that'll be coming out, a number of new tools underneath our NCP service that'll be coming out. We as a company, frankly, we're all in on AI. We want to build AI for our customers and really be the leader in in terms of bringing that capability out to the market. Frankly, internally, we also focus very heavily on AI. So when you see what our software development lifecycle looks like today within my organization, we have a lot of agents writing software and testing software and creating specifications and doing research. So, my mantra is to you know kind of raise the bar in both of those areas, to get better internally at delivering software with agents and to deliver agents and AI technology out to the marketplace. So what you'll see from us is a lot more of that on both fronts, and I'm really kind of proud and love to talk about both sides of that because I think they're both really interesting problems that are somewhat overlapping. Um but it's been uh it's been a really interesting and sort of challenging world that's been fun to see the results of. Um when you think about where we're going uh more holistically, I think in the industry, uh I think the people that are gonna be successful are the people who, as I mentioned a little bit earlier, are starting to kind of encode um their workflows and their and their knowledge of their industry, whether it's insurance or not. They're encoding that knowledge of how people work into artifacts that the agents can then use to do that work. And I think that's incredibly important because you got to start there with something basic, and then you'll continue to get more and more sophisticated into more and more complex parts of the organization, more and more complex workflows. And so I think the people that are doing that will see, I guess, more value uh from the AI world. And um whatever workflow that is, it can be AI enabled at this point. We're not finding much that we can't at least apply some AI to. I think the human in the loop is still important, especially in these regulated markets. I don't want to say we don't necessarily see that as a downfall. If having a human in the loop is what's necessary to make sure that you're maintaining compliance and related things, like that's an okay thing. Uh, we're still making a lot of progress and having a big impact with uh with the AI on a given workflow anyway. So, you know, when you look out there in the industry, I guess that's kind of my comment is uh I think all things could be AI affine, if that's the way to say it. And I think the people that are really building in and leaning into creating agents and creating the artifacts that those agents need are the ones that I think will continue to kind of just roll downhill faster than others, if that makes sense.

SPEAKER_00 19:04

Totally. Well, brilliant update. Thanks for all the insight and info and insider's perspective. Uh learned a lot and uh look forward to an update at some point in the future.

SPEAKER_01 19:12

That's great. And I really appreciate the time to chat with you today. I enjoyed it.

SPEAKER_00 19:16

Yeah, likewise.

Final Takeaways And Closing

SPEAKER_00 19:17

And thanks everyone for listening, watching. Also, check out our TV show, techimpact.tv on the Bloomberg and Fox Business Monthly. Thanks, everyone. Thanks, Doug. Thank you.