How a Serial CDAO Scales AI in Insurance with Verisk

0:00The more data that you bring to the equation, the more elements that you have in the algorithm. The higher level of accuracy, you should be able to reach with your outcomes. Hi, I'm Cindy Housen, host of The Data Chief. If you've heard we talk about data and AI before, you know that I believe everyone, not just data teams, should be able to get the insights they trust. That's why I'm proud to say that Thoughtspots sponsors this podcast. Thoughtspots' authentic analytics platform lets you simply ask a question in natural language and get clear, governed answers right when you need them. No fuss, no waiting. It's why companies like Cisco, Lyft, Hyatt, and Roach count on Thoughtspot.

1:01See what the future of analytics feels like at Thoughtspot.com. Welcome to The Data Chief. I'm your host, Cindy Housen. Today, we're joined by Louis Dima Digno, global chief data officer at Barrisk, the number one data provider for all things insurance and the leading strategic analytics partner to the global insurance industry. Louis is one of those serial CDAOs with leadership positions at organizations such as Hartford Steam Boiler, the Hartford and Axa US. In this episode, we'll explore how to build AI-ready foundations, the growing role of unstructured data in decision making, and why leaders must design systems that keep humans in the loop. Louis, welcome to The Data Chief.

2:02Thank you so much, Cindy, for having me today. Yeah, and where in the world are you joining us from today, Louis? So I'm just outside of Snowly Hartford in South Windsor, Connecticut. Okay, how many inches did you guys get? I think we got about 18 inches here. Oh gosh, it's incredible. There's like no place to put it. Hopefully by the time we air this, everyone will be in spring mode. And for those of you watching us on YouTube, Louis, the easiest way to spot you at a conference is the iconic bow tie, share with us. Yeah, definitely. So I became the bow tie data guy when I hung up my Air Force uniform. And so being in the Air Force in the military, obviously it's very easy to know what you're going to wear day in and day out. And so when I came over to the commercial side, I adopted the bow tie as my signature and it became my everyday occurrence.

3:02And I did an inventory around Christmas time. I think I have 227 bow ties now. So my challenge to everyone is, if you ever see me with the same bow tie twice, I owe you a coffee. Oh, there you go. Well, it definitely is a classy look. And I should say, thank you for serving. Now, you mentioned serving in the Air Force, but you were actually a retired colonel. Is that right? That is correct. You had 27 years total, federal time, 12 years active duty, living all over the world. And then 15 years as a reservist, which really took me into some really unique areas I had the opportunity of working at the Pentagon and was deployed from there to a number of initiatives that needed a specific process improvement capability. And I'm thinking about going from being a colonel

4:06in the Air Force, high pressure. I'm going to assume at one point you were looking after 22,000 aircraft. So a lot of data, but going from that to heads of data, to give me the connection or disconnection here. I have to say, in preparing for our discussion today, I went back and I started really looking at the connectivity between my military service and where I am now in the data world. And one of the things that really resonated with me and doing that review is looking at the Air Force's core values. And the core values, the three core values of the Air Force really align so nicely with what I do in data every single day. And so the three core values of the Air Force are integrity first, right? Which again is a big piece of, we need to make sure that we've got very accurate

5:08and high quality data, service before self. And again, we are really focused on the aspect of what are we doing with that data and being able to use it to make informed decisions. And then excellence in all we do. And that component of really continually trying to improve the data, improve the at level of accuracy and improve the level of quality, it influences so much of the confidence associated with the outcomes of the data. And so again, those core values of the Air Force, I think just very much aligned with being a data professional being someone who focuses on data and the foundation of our decision-making capabilities throughout the organizations. I think those core values really align nicely. And I think it's probably a big reason why I am in the data space today.

6:09I did focus for a while on data science and the statistics side of the equation, but I found that the preparation and really getting the information ready for data science, there's just so much opportunities for efficiency there. And as I said, I saw the alignment with the Air Force's core values. I like the way you connected those core values to the role and I wanna pull on one of the threats to start with and that's service. And if you think about the role of a good offensive chief data and analytics officer, they're more a collaborator rather than, let's say some old school leaders would be more empire building. How have you seen this show up? And also if you maybe think about some of the CDOs that you talked to earlier in their career,

7:10how does that play out? I think that again, looking at it from a standpoint of being a team player, right? So that whole idea of service before self of really trying to focus the aspect of, what is it that we're doing to prepare the data for analysis, for experimentation, for intelligence? And being able to be ready when the data's necessary. And so that aspect of it of really doing the preparation, really trying to get into an area where data engineering in the past has really been the focus on this area. And I think that with a lot of the tools that I've seen through my career, I have really changed it so that the data engineering aspect has become a lot more automated, a lot more opportunities through entity resolution, master data management,

8:12a lot of different tools out there that allow us to get the data in a better situation so that it is ready for the data scientists, for analytics, for intelligence, layers to be added on top of it. And so that preparation of constantly improving the data and really trying to be ahead of the game so that when the questions come in, you're ready to be able to participate. As I say, it's like a team sport. You're a big part of that preparation aspect of trying to have the data ready and available for others to be able to utilize it. It is a team sport. I think that's a great way to phrase it. You're messing with Randy Beane's industry survey for average 10 years for CDAO. So why do you change your teams so much? So I gotta say that the opportunities that I've had

9:14in front of me around changing from companies that the timing has been really good is what I'll say. And so in each case, I really feel like each organization I've left in a much better situation than when I got there. Being able to put governance in place, being able to have measurements around accuracy and levels of quality, being able to bring together data through a master data management or entity resolution. Always efforts for me have culminated with each of these organizations as I effectively finished what I started and was able to walk away from each of those organizations, leaving a strong governance aspect, leaving a strong organization that was able to pick up the ball and continue to move it forward. And for me, a lot of it has been around, you know, how am I moving with my family as well

10:20to ensure that, you know, I've got opportunities to be, you know, balancing my work life appropriately in each one of those situations. And so, you know, I think that to Randy's point, you know, he says about two years, two and a half years, I think is the threshold. The good thing is is that I've lasted more than the two and a half years in each one of my situations, which again, I feel very strongly about that, again, from a contribution standpoint, but it also was a, like I say, an opportunity for the next generation of leaders within that organization to step up and to take the reins and continue to move the ball forward. Yeah, and I do see that some move on more quickly or that quickly, either because they get frustrated or they get pushed out

11:21because they're driving change too quickly or as you described, it's mission accomplished. And now I'm ready to move on. If I can make an observation to you though, given that I first met you at Hartford Steam Boiler, if I remember correctly, I think it's also that you are scaling your impact in the industry. If I look at how you went from, let's say, one carrier and the internal deployment to now at VARISC, you're really serving all carriers and regulators and external deployment. So it's almost like scaling your impact. Do you agree or disagree? I would agree, and I think that's one of the main reasons where I did join VARISC is I saw the opportunity to take my learnings and my experience. I was effectively one of the first CDOs in the insurance base back in 2012.

12:22And so being able to get to this point in my career, where I do have an opportunity to influence an impact across just about every carrier out there. It really has made a great opportunity to leverage my experiences and to bring my knowledge base to a much broader audience and to help the rest of the insurance industry to leverage this data as we continue to optimize and influence, especially with all the opportunities within the AI space now. Yeah, so for those who are not familiar with the insurance industry, describe a little bit more precisely what VARISC actually provides. Sure, so we focus a large portion of our effort is around forms, rules and lost costs in the insurance space.

13:23And so that is really about how are you building your program within your insurance carrier space? So having the forms, it has the opportunity to reduce litigation costs and utilize the standardization that we've been able to bring together with all that information, leveraging the rules across all of the different statutory regulatory environments that manage the insurance industry, especially in the United States, and then leveraging the lost costs to ensure that you're pricing appropriately and predicting claims well within your organization. And so that's the foundation of what VARISC does is being able to support across the different carriers in those three areas of forms, rules and lost costs.

14:27But then we also bring together a lot of information around antefroth and claims support. So we bring together all the claims across the insurance industry. And by doing that, we have the opportunity to look across a number of instances to determine whether there are fraud issues happening there and obviously, or not obviously, but one of the biggest impacts within the insurance space is detecting fraud and being able to identify that early so that we can keep the costs down for the policy orders out there. Yeah, so let's make it real for everyone. We have these extreme weather events, whether it's in California and the West Coast, the floods on the East Coast, it's snow, ice damming on roofs, a tree hits a car. So it starts with the policy, the underwriting. Should I ensure that property or that business?

15:29And then it goes through the claim, where does your data come in now? What happens? So we're across the entire value stream for the insurer. We help, as I said, in helping them determine what products they should be in and being offering to their constituents. We also work very closely with regulators across 50 plus jurisdictions. So we're continually giving states and other levels of jurisdictions information about carriers within their environments within their localities so that again, they can see that there is no bias in our pricing or in the carriers pricing. There is no impact that's negatively affecting the policy holders in looking at things like catastrophes, like the wildfires, the flooding, the snow.

16:31All these things are things that we are on top of from a prediction standpoint. So we have a catastrophe and risk solutions organization that very much focuses around what is happening in those spaces and helping to determine what the level of risk is associated with some of these events happening and the cost impact associated with them. And then from that, you go into the, as I said, the forms, rules and loss costs of building products. Then you have the underwriting and pricing aspect of ensuring that you're pricing it appropriately. And then you have the follow-on of the claims of determining what is the impact of the claims. Are they being serviced quickly? All through that value stream, the entire insurance value stream, we are benchmarking, right? So we're bringing all this information together from the carriers that contribute to us. And we bring that information.

17:31We can share that back with the organization so that they can see how they're performing compared to their competitors as well. And in doing that, then they can go ahead and they can set their prices and determine what areas of the industry they want to be impactful in. Okay, so thank you for taking me through that. So now you and I have talked about AI-ready data and you've talked about master data, but you don't own a lot of this data. You're aggregating synthesizing it. So first, give us an idea of the breadth. How many data sources, what data volumes are we talking about here? So we're talking over 40 petabytes worth of data. And so again, this is longitudinal data. So we're talking about over time. We're collecting data from carriers that contribute the data to us. And so by them contributing to us,

18:33we have the opportunity to go ahead and bring all that data together from an entity resolved standpoint, whether it's at an individual level or a commercial level. We can see all that information. We can share that back with the companies that are contributing to us. And so given that you don't control the processes at the carriers and how the data really arrives, how do you ensure that the data is complete, consistent, clean enough to be directionally accurate? So big part of what we do is we look at consistency of the data over time. And so again, if we get a data set from a carrier, we'll go ahead and we'll look at the previous time that they had gone ahead and sent it to us. And then we'll do some comparative studies to ensure that we don't see anything that looks like noise through the signal that we would expect. And so we are doing that each time we get another data set

19:36from a carrier to determine whether it is in line with what the expectations are. If it's not, then we work with them around trying to either correct the data or to ensure that what we are seeing is accurate. So there's that part of it. The other part is the third-party data that we go ahead and we bring in from other vendors. Third-party data aggregation as well is something that we go ahead and we fortify our data with to help give us even more insights into specific areas and to help ensure that the accuracy associated with the data that we've gotten from the carriers is also in line with what we expect. Okay, so you have a lot of data. You make it into clean, usable data for benchmarking, for product development. Now let's talk about how you get it in the hands of decision makers. Take us through how you got to an AI first data strategy

20:37and if you will, the role that ThoughtSpot has played here. So a large portion of what we have been looking at over the last couple of years around AI is, how do we go ahead and embed AI within our environment? And again, we've got over 150 different products out there and again, it goes across the entire value stream of the insurance offerings. So there's a number of different ways that we could look at this. We could go ahead and we could build AI specifically into our products to help through different agent features associated with that. We've got ways that we could maybe even bolt on AI from a searcher or summarizations capability, again, a different level of the agent aspect of an offering.

21:37But then we've also got the opportunities where we've partnered with ThoughtSpot that brings AI embedded within their product. And so by having our data available through the data sets that we go ahead and populate through the ThoughtSpot products, we've got the opportunity to utilize the spotter and the natural language processing capabilities to interact with the data. So that you can, I like to say you can talk with your data. Yeah. And I really feel strongly about that. I've been using ThoughtSpot through a number of the organizations that I've been with for. I guess upwards of, it's got to be close to 13 or 14 years now. And again, I've always been enamored from the standpoint of having that Google search type bar to start with, to interact with your data

22:38is really just, I feel so easy for someone that's being exposed to the product for the first time. It's a natural way of interacting with the application. Yeah, so maybe take it a step further and make it real. Can you give me an example of a type of question that a carrier might ask of this data? So areas that we've been really focusing on with ThoughtSpot right now have been the regulators. And so of the 50 plus jurisdictions that are out there, we've gone ahead and we've created products on the regulator side that we've gotten into the hands of regulators where they can start asking questions specifically about concentrations of risk within their localities. And so they can really start to drill down into areas to start asking about specific industries, specific coverages, types of claims.

23:38All that information now is something that a regulator who may not necessarily be educated in all the latest technologies out there, right? And they may not be up on all the latest AI capabilities. But with that ThoughtSpot interface, they have the opportunity to interact with their data and the underlying aspect of what ThoughtSpot's delivering to them is our insights that not only give them their answer but also give them follow-on questions or insights that maybe they weren't looking specifically for. And that to me is really one of the big benefits of ThoughtSpot is that it's trying to anticipate or at least see what makes up the answer of what it is that you're asking for. And so it gives you that opportunity to drill down and really understand what's driving. Yeah, one of the things that I saw at the Vic conference last spring was how quickly your teams

24:40could look at the way the tariff impacted repairs. So it was something like the cost of nails imported from this country and how it impacts roof repairs, which just kind of blew me away. Well, we've been using ThoughtSpot quite a bit on our exactimate suite of products. And again, the exactimate products are really about helping contractors determine repair costs and understanding what are the impacts of those costs, especially from that locality. And so you can imagine, as you go across the United States, there are different prices for different items in things such as wood, nails, tear point, insulation, all those types of things that come together to help repair from a claim from an insurer.

25:42And so having a tool like that and having the underlying understanding of what's driving some of those costs really helped to give the contractors some insight as to what they should be preparing for. Yeah. So Lewis, I want to come back to another point that you raised. And this is the dark side of AI and deep fakes. And this is something that scares everyone, but if I think in particular financial services and insurance, I may get a text this morning saying, you are pre-authorized for a reimbursement for ICE damage. Or you may, a carrier may receive a photo that is a deep fake to damage. It's not real. How are you battling this? What are some best practices leaders can learn from you? Well, a large portion of what we do on the unstructured side

26:46of the house, specifically with pictures and videos is really trying to understand, are they AI generated? Are they being utilized from maybe another location on the web where someone goes ahead and takes a picture from someplace else on the web and submits that as part of their claim? And so what we've been doing, we've been working with a number of vendors to really try to understand how do we identify these types of pictures and videos that are being submitted as to whether they are deep fake AI generated or whether they are truth. And at the end of the day, our emphasis is really emphasizing how can we get to the truth and really understand the fact base of what it is that we are evaluating or what it is that we are sharing with other organizations,

27:48other carriers, around what to be aware of in these spaces. So we're constantly learning through this AI education. There's so much opportunity to utilize the AI for good. But as you say, there's definitely a nefarious element out there as well that for everything that's good associated with AI, somebody's going to try to figure out what bad thing can they do with it as well. So we're constantly trying to stay on top of that. As I said, we're working with a number of vendors to identify areas where this is being implicated. And again, we're working with politicians and regulators as well to help understand how do we earmark AI generated

28:49pictures and video to have something embedded into them so that we can identify them as AI generated. So we're working through the regulators and through the politicians as well to help try to put some of these efforts in place to protect the carriers going forward and obviously that goes on to the policyholders. Yeah, so these deep fakes pose a risk, but also if we think more broadly access to unstructured data can be game changing. As you think about the range of unstructured data sources, whether it's documents and drives, images, voice, PDFs, with best practices or policies, what sources do you think will be the most game changing, not just for financial services, but also or not just for insurance,

29:50but as you've spent time in financial services too, what do you think the possibilities are here? It is truly amazing. I think when you go across all the different types of data that can be captured. And so when you go ahead and you really start to look at where we started, everything was so rose in columns, number focused, very structured, very fact based, very mathematically oriented. And so now to take these math foundational components and start to put them into this unstructured space, it really starts to get into an area where there's a lot of opportunity for manipulation and areas of, as I said earlier, the bad guys to exploit.

30:53But at the same time, it expands the amount of data that is being brought together that, in my opinion, the more data that you bring to the equation, the more elements that you have in the algorithm, the higher level of accuracy you should be able to reach with your outcomes. Totally. So as we bring that more unstructured data sources together, as we create better data models around graph databases or knowledge databases, vector databases, all these different capabilities now of bringing this information together and to have orientation on that data to have the proximity of the vectors help to understand what are the similarities associated with the data. It really starts to expand your mind

31:54as far as what are the capabilities in influencing the outcomes of what it is that you're trying to understand. And so I keep going back to, hey, your confidence in your modeling has no place to go but up the more data that you go and add to the equation. Let me challenge you here, Lewis. So I am super excited about being able to access unstructured data and combine it with our structured data. But how clean is this data? Have we followed the same data management and data governance principles on unstructured data as we have with our transactional system data? Well, I think there's number of challenges associated with it. And again, we could spend an afternoon talking about all the different capabilities associated with it.

32:56Where I see a number of areas that are interesting is through a lot of written documentation. And so when you go ahead and you start to look at the libraries of information that's out there in the insurance space, through policies, through contracts, through claims, all this information that's been written down or documented in some form or fashion, having that information now available to you again through a question-answer type effort to help the adjusters get to an answer faster, to help the policyholders recover from a loss quicker and more accurately. All of these aspects of the unstructured data being fed into the equation has the opportunity to again get you to a outcome faster

33:58because it's continually learning. It's continually taking all of these interactions and it's starting to see the patterns. It's starting to see the opportunities of how to bring this information together and get to an outcome faster for the policyholders. And it's in those areas that again, when you look at the frequencies of these things happening that are real, that are factual, that are truthful, those are the things that again, we're really looking to emphasize to be part of our data sets going forward. It's finding the manufactured components that don't look right, don't follow the trends, that don't fit the patterns. Those are the things that we're gonna go ahead and continually look at harder, scrub them out of the database, try to influence so that we've got better standards and understanding around I'll mention the dirty word,

35:03but governance, it's really about understanding, how is this data getting in there? Does it fit what we expect it to be? Are we using it appropriately? Do we keep the right controls around it as to who has access to it and what they're using it for? All these aspects are rules that we are continually trying to put up or guide rails that we're trying to put up so that people do understand that this is a fact-based that we really want to be protective of it, right? And we want to make sure that it stays unadulterated going forward. Yeah, it's funny, you called governance the dirty word. Somebody once said to me, governance is code for no. And if we think, go ahead. Well, what's funny is that again, over the 13, 14 years that I've been doing this specifically as a Chief Data Officer,

36:04I will say that governance has always been that barrier to progress. And what I would like to say is, I've tried to move away from using the word governance as much as I like to use the word observability, because I really think observability shows more aspects of what it is that we are doing with the data. And by understanding what's happening, having that intervisibility of what you're using the data for, how much it costs, where it's being stored, all these features associated with the data that will really help you to optimize your environment, it's an observability piece that expands governance into a standpoint of not only do I know what I'm using the data for, but I'm knowing how I'm using the data and with what I'm using the data. And so there's so many other features associated with it, it really gets down to the who,

37:04who, what, why, where, and when associated with the data as opposed to just the know. Yeah, yeah. And I like the way, so another industry leader, Paul Drennan, he talks about getting people to team, yes, from team know to team, yes. And so that observability or enablement is a good way to frame it. So as you think about Lewis, now you have this, you have more data, you have more context, you have AI helping speed the time to insights and ultimately outcomes. What is your view on when to have human in the loop here? Well, I think that again, from the standpoint of insurance being as regulated as it is, and to have as many statutory functions out there that are looking across all these different areas, one state may go ahead and find an issue today

38:06and all of a sudden a proliferates to a number of other states. And so it's interesting to have the level of scrutiny that is constantly being looked at across the insurance sector, the insurance industry. And again, I would say that different states have different priorities around what it is that they're focused on for their constituents, right? So Florida has a very, very different PNC standpoint when you go ahead and you look at the impact of hurricanes and flooding issues associated with their constituents as opposed to somebody in middle Oklahoma who has more wind storm issues associated with what they're doing. And so our tornado, I guess, aspects as opposed to hurricanes. And so they affect their constituents different. They, you know, there's different aspects

39:08of how they're being evaluated. And I do think that this AI capability, especially through tools like ThoughtSpot are going to give insight that just wasn't there before. Because again, it's asking additional questions that maybe the regulators weren't specifically thinking about. But as you drill down and as you ask different questions, different ways, you're going to get insights that maybe you weren't expecting. And you'll see, you know, nuggets of information that maybe didn't have access to before. Yeah, yeah. Well, Lewis, we've covered a lot of ground from what makes a good CDAO to what is Varisk do, how the insurance data industry is changing, getting AI-ready data, AI fraud and human in the loop. Let's shift to a fun lightning round. What do you do for joy outside of work?

40:11So I'm a beekeeper when I'm not handling data. And actually, it has quite a data aspect to it. I do have some internet of things, tools associated with a couple of my beehives. So I am collecting data on them as well to understand their health and their production. But it is definitely a very, very different aspect of getting out there and working with nature and having the opportunity to deal with bees and their communities. That is different. Now, when you were visiting us in the Mountain View office, did we show you the beehive on the roof? I was told about it. I unfortunately didn't have the opportunity to go off, but yeah. Okay, I need to get you some honey from there. Just cool beehives on the top of the tallest building on Castro Street. Who is someone famous or not that really inspires you?

41:13So I have to say, and again, I don't mean this to be self-serving, but I will go back and say people in the military. And I'm going to say that in a general standpoint because again, the service aspect of that, the military aspect, the public servants that are out there are firefighters, are police officers. They're the folks that inspire me every single day. And because they are out there trying to serve us and make our lives more safe and being able to put their lives on the line for us to make our lives better. Yeah, and beautifully said, without our freedom or our safety, what do we really have? How about a book or a podcast that has significantly influenced your thinking?

42:16So I did, so I drive back and forth to the Jersey City office each week. And so I have a lot of time to listen to books on tape. And most recently, I listened to, if anyone builds it, everyone dies. And so again, it's all about superhuman AI and it's quite interesting. I think it gets a little overboard and preachy at times, but it does show the impact of some of the things that are happening in this AI space. And I think that also when you go ahead and you add into the equation, the capability of quantum computing that's coming down the road. And again, it matters which labs you talk to, but it sounds like it's a lot closer than I think most people really believe. I know that AWS and IBM have quantum computers that you can effectively run products on today

43:16or run models on today. But it's gonna be a lot more prevalent within the next five years. Yeah. And so again, having that capability of that level of processing associated with the advances of AI practically weekly, I think we are in for some really exciting times ahead. And I do think it's important, not only for the builders of AI, but for the buyers of AI to know what's possible and to buy ethically, really design ethically and buy ethically. So I'll have to listen to that on my next drive to the airport. Lewis, you can decide the last one depending on your mood in the moment. Either what are you most grateful for? Maybe be on the obvious of health and family or something that you are particularly proud of and accomplishment from the last week.

44:20Oh, from the last week. We do time box it. So here's, so our company has been going through a number of evaluations of AI vendors. And so this past weekend, I went out and I built myself my first app using one of the AI vendors. And as a matter of fact, what I did is I used three different AI vendors and I gave them all the same ask. And the amazing first off speed of response, the capability of the apps that they produced and the specificity of what I was asking versus what they created, I was just completely blown away. And again, I haven't coded in 20 years. And so to have this capability at my fingertips again, is just amazing.

45:21And I would encourage anyone, literally anyone to go out there and start experimenting with these. It is so easy to get into a space where you are literally creating applications that you would have had to purchase a couple of years ago in order to do what you wanted or you would have had to pay someone to go ahead and create for you. But the level of customization, the level of specificity of what they're capable of doing and the level of complexity and completeness was just, I was completely blown away. Yeah, what an amazing time, Lewis. Thank you so much for being on the data chief. Well, thank you so much for having me, Cindy. It's been a pleasure. And I look forward to speaking with you again soon. Like twice. If you enjoyed this conversation, please rate or like it on your favorite podcast platform. If you would like more inspiration and thought leadership,

46:24follow me on LinkedIn or visit thedatachief.com.