Translating AI Models into Business Value From Governance to Deployment - Thomas Holmes of Akur8
The AI in Business Podcast · 21 Jan 2026
Source: https://podcast.emerj.com/translating-ai-models-into-business-value-from-governance-to-deployment-thomas-holmes-of-akur8
Transcript: https://podcasts.opennash.com/shows/ai-in-business/translating-ai-models-into-business-value-from-governance-to-deployment-thomas-h-f8ffe7/ (via OpenNash Podcasts, machine-generated)

Welcome everyone to the AI in Business Podcast. I'm Matthew Demello, Editorial Director here at Emerge AI Research. Today's guest is Thomas Holmes, Chief Actuary of North America at Akur8. Akur8 provides software for non-life insurers to support pricing and reserving work for over 300 customers across over forty countries. Its platform helps actuarial teams build and manage pricing models and reserving projections with an emphasis on speed, transparency, and reliability. Thomas joins us on today's program to discuss what it takes for insurers to move from AI experimentation to enterprise deployment, especially when governance has to satisfy both executive stakeholders and regulators.

Our conversation also explores practical ways to leverage workflow with measurable ROI, including building quote unquote translation layers between actuarial, IT, and leadership teams, establishing a single source of truth for reducing rework and errors, and using opinionated, auditable processes that make it easier to scale adoption without sacrificing explainability. Today's episode is sponsored by Akur8, but first, are you driving AI transformation at your organization, or maybe you're guiding critical decisions on AI investments, strategy, or deployment? If so, the AI in Business Podcast wants to hear from you.

Each year, Emerge AI Research features hundreds of executive thought leaders, everyone from the CIO of Goldman Sachs to the head of AI at Raytheon and AI pioneers like Yoshua Bengio. With nearly a million annual listeners, AI in Business is the go-to destination for enterprise leaders navigating real-world AI adoption. You don't need to be an engineer or a technical expert to be on the show. If you're involved in AI implementation, decision making, or strategy within your company, this is your opportunity to share your insights with a global audience of your peers.

If you believe you can help other leaders move the needle on AI ROI, visit Emerge dot com and fill out our thought leader submission form. That's Emerge dot com and click on Be an Expert. You can also click the link in the description of today's show on your preferred podcast platform. That's Emerge dot com slash expert one. Again, that's E-M-E-R-J dot com slash expert one. Without further ado, here's our conversation with Thomas. Thomas, welcome back to the program.

It's a great pleasure having you.

Great to be here again.

Absolutely. We talked so much in our last program just about the actuarial space. We're seeing insurers move beyond experimentation. The real challenge is now operationalizing AI within governance frameworks that satisfy both regulators and executives. This is becoming an increasingly complex space. The focus is shifting from building models to implementing them with precision and ensuring that actuarial accuracy uh translates into financial performance. We see emerging tools, I I think in stiff competition for AI word of the year's tool creep or phrase of the year's tool creep, at least going from 2025 to 2026.

And such is the ones where you can produce. And they're now defining how insurers bridge the gap between actuarial modeling, IT execution, and leadership decision making. Just taking a couple steps back here, how are you seeing insurers really go about that unenviable tasks of having to appease both stakeholders as well as regulators? How should leaders think about operationalizing AI within their governance frameworks to satisfy both?

The good news is that there is alignment between these two frameworks. Uh, regulators in the end, they just want to understand what you're doing, why you're doing it, and that you did it right. Your stakeholders in your company really want the same thing, plus a little bit more of, does it solve our specific needs for the company itself? And in order to answer those questions, we really need to do some some translation between these different groups.

So data scientists, actuaries, leadership, regulatory, IT, they all speak very, very different languages. And so Akur8 is a French company with an Italian head of product that's growing in the United States. We have employees of over twenty five nationalities, and I would say the barriers to communication that we have internally are way smaller than between actuarial leadership and IT. We at least speak the same professional language. And so when you're doing this, you need to have something that performs this translation.

You need to have processes or software to actually help you make this transition. You need to have that. I don't even want to use the word handoff. I want to say translation of this item so that it's not taking the material from the actuaries and handing it to leadership. It's taking the materials from the actuaries, transforming it into something that uh resonates with leadership, handing that off. It's taking that rate plan, it's transforming it without effort into something that can be ingested by IT, by by a rating engine.

It's it's really that that transformation, that translation, rather than having a document with all the information, it's having it in an understandable format.

Absolutely. A question we're asking, especially a lot of BFSI folks writ large on the show about the regulatory space. Something I think everyone across even vast financial services has has come to grips with in the last few years is that relationship with with regulators changing, especially when I'm talking to our friends in banking, KYC, it is very much becoming, oh, the homework that, you know, the FBI used to have to do, they're expecting us to do that homework now, and they're just checking it.

And that relationship, uh, it not only changes, but I think a a huge force behind that, not just, you know, what what we see in kind of the the public sector is that it's hard enough to keep pace with the technology and regulators are doing their best. I'm sure this is this is front row center for a company like yours where you're operating both in the United States and Europe, just where the standards are so completely different, but everyone's struggling to keep up pace with the technology.

No one is succeeding because the technology is moving so fast, uh, regulators especially so for traditional reasons. But we end up with this gap where even for, you know, especially the insurance space where so much of it uh uh of the game has now been, oh, we need to get ahead of the regulators. We need to do a better job regulating ourselves than even that next round of regulation just to keep pace with customer expectations. Not what the world was like in the eighties, of course, but this is this is the new world.

And even with that dynamic, we see a gap of regulators are trying their best, but they're actually stuck in the past, not just maybe uh in in the sense of the uh the level of scrutiny, but actually that they're not being hard enough and they're using an aging framework that doesn't even serve business nor customer. Just wondering how you think about these gaps, and especially at a place where you're answering to two different standards of of AI regulations, both from the United States.

It's a little bit more piecemeal and and executive orders and full steam ahead on AI. And uh, the EU that has rights to privacy and is really trying to get their arms around this entire industry.

Yeah, that's a very good question. It has a lot of different facets. And before I dive in, I also want to share some differences in regulation in insurance. People might not, uh, know as much that the insurance regulation around pricing is very, very relaxed in Europe in general. While you can't use some of the same rating variables that you can in the U. S. obviously with the, uh, you brought up the right to privacy, credit score is obviously not something used in Europe.

Uh, but your models, for the most part, don't have to be filed. You just use them, you can update them them daily. Prices are incredibly less regulated. And so the types of models that you're using to price can be very, very complex i in Europe, and yet still they're trending back towards very transparent models for pricing. And the reason being, like you said, this this alignment between what the regulators expect, what the consumers expect, and then also what the insurance company, uh, needs.

And that is a complete understanding of what you're actually doing in the market. Insurance pricing is something that is very specific, and the implications for doing it wrong, the implications of changing those rates have very great impact on the company due to policyholder retention, dislocation, and how that affects your bottom line. So when we think about AI, we we really want to think about it in a couple ways. One of them is how you gather and put together information.

I think this is where there's a lot of freedom, a lot of usage to, uh, scrape information, get characteristics, get something that is easily understandable, and the key is to make it stable. Once it is stable, that is the that is the key to using it in insurance. And then when you deploy your rates, having that transparent rate is going to prevent you from having a lot of headaches with regulators, internal stakeholders, and customers. When you are inputting inputting certain types of models that might be explainable i in aggregate, but not locally, the first time you input it, it might be okay, it might have some some weird results.

The second time you put it in, you refit your models. And what that means is that you're going to have some people that have impacts that you just can't explain. They're going to be very hard to say, why did this person get 200% rate increase? And the answer is the models did something different, they found some different niche spot. And for something like a vastly different Netflix recommendation, that's fine. I think that's that's just what whatever.

I I don't really care if this person is now getting recommended horror movies that maybe they care. It's it's not very, very good. But but

Variety is the spice of life, right?

exactly, but but the consequence for that I is quite small. The consequence of changing prices that much without being able to fully explain it is in fact pretty huge. It's and this isn't just a regulatory question, it is a reputation question, it is a company question, it is a question of what is your dislocation going to look like? And everyone insurance knows that the wider your dislocation, the more people leave, and retention is gold for insurers.

The longer you you keep the the best policyholders, the the better. So choose your AI wisely. There are going to be some aspects where you can use the most black box thing ever, and the consequence of getting it things wrong is is pretty slim because what matters is the aggregate prediction. When it comes to these high consequence matters, you need an AI that outputs something that is extremely transparent.

Y- you're already a little bit answering my next question, but maybe we can dive in a little bit harder. I I think maybe the most difficult point we hear from the audience in terms of really thresholds is understanding what's that difference between knowing we've just been playing in the sandbox with an important problem. I I think the operative word ends up becoming you want to go to where people are drowning, because that situation's not getting any better.

You can experiment, you have only one way to go b up. It's important because you have people, you don't want people to drown over something useless. It's important enough that you're going to be invested in in the stakeholders. So, all right, we got we got a good place where we know where to experiment. How do insurance leaders know when it's safe to be moving from that experimentation phase with AI deployments to that enterprise-ready deployment that will need KPIs need to be integrated into the quarterly uh thinking and have a much closer relationship to leadership and the stakeholders.

Great question. Once you've been playing around in this space, you're going to get to know the risks associated with a given implementation. You're going to get to know the desired outcome. You're gonna get to know where you can kind of go off the rails and start getting some some odd results. And the idea is to put together some guardrails around the process. In the last episode, I I said an opinionated framework, and I'm gonna come back to that and say that once you've done your experimenting, you need to say, this is how we use this AI to solve this problem.

And now this is going to remove some of your immediate flexibility. You can't do everything in the world, which maybe to a a single power user sitting at their desk, maybe they they want to to do something special. But when you're thinking about a enterprise-wide rollout, you need a very set structure, again, that can be changed at any given step. You need the ability to deviate it from, but in general, there has to be a default way to answer this problem using AI that you can roll out.

And what that is is an opinionated structure saying, here is the first step. Here is all of our checks that we provide. Here is the second step. Here are all the checks you can provide. Here's how you edit it. Here is the third step. And by going through this process and setting it up, you have you have two things. One, you have a framework for usage. Two, you inherently get a framework for governance. And so by setting up that structure a and setting those those guidelines around how you use it, you can really answer both of those questions.

And this is very important because there is no single AI governance framework. There just simply isn't there. There won't be it. It's evolving too fast. I I got I went through some questionnaires in my past where it said, answer the these these hundred questions about your model. And about 60% of them were were NA. They just didn't apply. And there's gonna be a different 60% that's NA for different types of models. And there's also gonna be some things that weren't on that checklist that are unique to what I'm doing.

And so what you need is that that framework, and you need that that experts to tell you here are the things that can go wrong. The outcome of this process is also it should be very difficult to go through this process, come out with something bad and not know it. That is really the key to putting together the this process is at the end of it, because of your guardrails, because of your opinionated framework, you should know if something has gone wrong.

Yeah, I I think, especially for our cross-industry l- listeners, that opinionated framework, that flexibility is gonna have large applications e- even outside of insurance. Just before we're moving on to trade-offs in in uh where we wanna go with this conversation, just wanna get a sense of how do you define opinionated frameworks and where they're best deployed to really drive AI deployments, maybe maybe regardless of space in the enterprise. Good question. So opinionated frameworks are based on your industry.

So there are various ways to to build a model. There are various ways to, uh, analyze differences between these two models, but in insurance specifically, there are ways to do this that are actuarially sound, and there are ways to do this that that are not. And you set these guardrails, for example, I in modeling, you you wanna do variable selection. You set your guardrails. Here's the process that we're going to do for variable selection. Here is the things that we want to consider.

Here are the things that you look at when you when you are, uh, fitting a model. And it can be also as simple as here are the types of models that are even in scope. And so a lot of it comes down to what is an actuarially sound model, what is an acceptable level of transparency. And then when it gets to the, uh, rating, it's also what do we need to explain to the regulators when we say do it right, when we say make sure that nothing has gone wrong.

There are very specific questions in there that insurance companies will want to answer, and that is what I mean by the opinionated framework, is that it is customized to your industry and not just throwing out this this general algorithm for for use. Absolutely. And I think it's been so much part of the conversation in 2025. I I though I think it's evolved. Back back two years ago when chat GPT and LLMs were fresh in the public's mind, I think, especially at the forefront of AI, the dynamic or the question was build versus buy.

Which one are you doing? Very binary. And now, no, that's actually it's really, hey, what's your ratio? How much are you building versus how much you're borrowing versus how much you're buying versus, you know, maybe you got some, uh, hybrid kind of systems in between. How do you think about those trade-offs or really answering the question of what that ratio should be between building AI tools in-house and buying proven solutions, especially around actuarial modeling? The real question is around return on investment and what types of resources you are using to build.

As an actuary, I have built tools. I am not a very good software engineer. Uh, actuarial salaries are not cheap. And we spent a lot of time building these these tools that were okay, but didn't really stand the the test of time. And that's because maintenance costs and this technical debt are just criminally underrated. Uh, when you say we can build it in-house, how many head count are you spending to recreate something that that exists? How does that actually compare to to your price?

And then infrastructure, if you're building a a SaaS solution or if you're buying a SaaS solution that's taken care of, are you now including IT in that? Are you including your technical leader that gets pulled into all of the questions when the tool inevitably breaks a a couple times? There's just a lot of in-house commitment a and problems. And I would say there are things to build in-house, and those should be the things that are proprietary and highly differentiated.

So I haven't seen anyone rebuild Excel. Everyone uses it. It's a huge cost. Uh, and yet for some reason, I haven't seen anyone try a and reprogram it. Why is there? Because there's a good solution that does it, and you can do better things with your time than reprogram Excel. And I think that is really the, uh, a huge key when you are thinking about building versus buying is, am I building something that is out on the market today, or am I building something that is actually going to give me a competitive advantage?

I think a lot of this also comes from a general fear of working with vendors. Like, are they going to live up to their promise? Are they going to fully deliver? Can I I trust them? And I would say that is a a relationship that you can feel out. See how the support is, see how thorough the the representatives are at this company, see about their knowledge base, see how well they understand your very specific problems.

Uh, build that relationship out, and then you can feel a lot more comfortable with these vendors rather than not buying due to fear. You can fully evaluate and make sure that this is a company that will solve your problems, that will evolve with you, that will provide you the level of support that you need, uh, to succeed. Absolutely. I think a lot of build versus buy conversations tend to center around those gaps between actuarial, at least for the insurance space, between actuarial and and IT.

Uh, you know, there are equivalents in other industries. I think it's a very popular space on the vendor side to say, hey, hey, you know, in in the stilt technology I was talking about in in the previous episode where, oh, you don't have to worry about your modernization. We're just gonna build on top, you know? And that can often be kind of the pitched, you know, solution for closing that gap between modeling, you know, in a business problem that's focused on and where IT needs to catch up.

How do you recommend, you know, insurance leaders think about bridging that gap between actuarial modeling and IT implementation? That's a very good question, and that's something that we've spent a lot of time thinking about at Akur8. And we realize that there are a lot of different areas that are your source of truth. When I go to an actuary and ask them, Where is your rating plan? Well, it's an Excel. Uh, well, we filed it. It's in our rating engine.

Well, we coded out a re-rater in in Python. The answer is there is not a source of truth, and that is problematic. Um, because actuaries have to perform re-rating in s in some way for the measurement process. I in the United States, maybe it's a Excel rater with a huge, enormous VBA macro. Maybe you've coded out in Python. Maybe you try to coerce your your rating engine into doing the this for you. But the answer is that there's really not one source of truth, and there there needs to be.

There needs to be one source of truth that that is used for both the actuarial re-rating and that is the object that should also be used in in production to program out these rate. Or sorry, there should be one source of truth to pull into your actuarial re-rating and also to push into your rating engine for production. This idea of of taking the actuarial rates and then reprogramming them is an unnecessary redundancy and it's a huge pain point.

It's a huge source of error because there's not only a a handoff, there's re-implementation of of rating logic, and there's really no way, or there's really no reason that you shouldn't have that single source of truth be used both for the actuarial side and for the the rating side. That's not to say that IT is is out of the the the cycle. IT will obviously always be in in implementation. There's always testing to do, but that core center of the product, that that rate order of calculation, where your rate tables lie, how they fit together, that is something that the actuaries should have visibility on and ownership the whole way through so they can own the part that they should

own. IT is on either side of that, is on either side of, like, well, here's my data coming in for my quote. Here's my data after the fact that I'm gonna send back to them. But that rating order of calculation can, and I think should be owned by the actuaries the entire way through the process.

absolutely. And, uh, I think that that y- you're leaning a little bit more into, hey, you need that single source of truth, but it needs to be governed, uh, correctly. Great if you have this great big tub of data that ju- that that you know is the truth. But if it that truth is not getting to the people who are needing it and asking the questions, then it's just gibberish, right? So how are you, or how would insurance leaders know they're being successful at the AI governance side from taking that single source of truth and making sure it's going where it belongs?

Well, the first answer is that when you have this source of truth, you don't have multiple versions of your source of truth anymore. You have the the one. And so that really simplifies your governance from the start is that you do have your single source of truth of this is my set of rates. And then that also allows you to really synchronize your IT and actuarial sources of of truth. There is one framework that's being used for both.

It's designed to flow through your whole process, and you are able to know that this is the object I used in my scenario analysis. This is the object that that I approved. This is the object that is in my rating engine. It's the same thing. And because you don't have to maintain your governance and versioning structures in all of these different areas, it's simplified.

And e- uh, especially in that way, you know, like d not take knowing when not to take that simplification, uh, for granted or know it's, you know, a a, as the phrase goes, too good to be true. Well, sometimes simple a and this can be so hard because, yeah, you you also have that a part of evidence that's like, well, you know, the simple answer is usually skirting over the truth. You know, sometimes when it's easier, it's easier for a reason, but it's always important to know when we can have that faith, when we can have that trust in the simplicity.

And hearing from experts like you really helps the audience to understand that. Thomas, really appreciate you being on the show, uh, for these last two episodes and shedding light here. Thanks so much for being with us.

It's great to be here, Matthew.

Wrapping up today's episode, I think there were at least three critical takeaways for insurance executives, actuarial leaders, and IT leaders responsible for pricing modernization from our conversation today with Thomas Holmes, Chief Actuary of North America at Akur8. First, the center of gravity is shifting from model building to organizational alignment, making sure technical work can be understood, trusted, and acted on across the business. Second, in high consequence domains, AI success is defined less by sophistication and more by confidence.

Leaders and approaches that stay stable over time can be explained at the level where decisions are made. Finally, return on investment comes from reducing friction in how work moves through the enterprise. When processes are clearer and information is easier to carry from analysis to implementation, teams spend less time translating, redoing, and reconciling results. Finally, return on investment comes from reducing friction in how work moves through the enterprise. When processes are clearer and information is easier to carry from analysis to implementation, teams spend less time translating, redoing, and reconciling results.

Interested in putting your AI product in front of household names in the Fortune 500? Connect directly with enterprise leaders at market leading companies. Emerge can position your brand where enterprise decision makers turn for insight, research, and guidance. Visit emerge dot com slash sponsor for more information. Again, that's E M E R J dot com slash S P O N S O R. If you enjoyed or benefited from the insights of today's episode, consider leaving us a review on Apple Podcasts and let us know what you learned, found helpful, or just like most about the show.

Also, don't forget to follow us on X, formerly known as Twitter, at Emerge, and that's spelled again, E M E R J, as well as our LinkedIn page. I'm your host, at least for today, Matthew Demello, Editorial Director here at Emerge AI Research. On behalf of Daniel Fagella, our CEO and head of research, as well as the rest of the team here at Emerge, thanks so much for joining us today, and we'll catch you next time on the AI in Business Podcast.