Third Bridge专家访谈称Verisk保险数据壁垒深、AI更偏增量而非颠覆

[[00:01:27]]
Could you share an overview of the Verisk Analytics business, including the evolution that the business has undergone over the last several years as it sought to become the pure-play insurance data and analytics provider that it is today?

[[00:01:46]]
Certainly. At its core was the ISO brand. When the company went public, ISO Inc still existed, but under the Verisk umbrella, and there were other business units that included risk mitigation, global risk. There was a healthcare segment and other financial services products. There was a push from the shareholder base that they thought Verisk was stronger as a pure insurance provider or data provider. The ancillary businesses were spun off. At its core came the Verisk core line services, which is really the fuel for a lot of insurance companies to select their rates. We provide the loss cost foundation that companies would then apply their own experience level to, and come up with their own rates. At the base of this engine is a proprietary database that has… We've been collecting data since the '80s or actually, I think, in the '70s, it started, collecting data across personal and commercial lines, home, auto, property. That data supports not only the loss cost, rules and filing, but also a regulatory need that the industry has.

Insurance is highly regulated by the 50 states, and their companies have a duty to report their experience to the regulators. It goes through a stat agent function, and there are other stat agents. Verisk is not the only stat agent, but we collect that data. We collect it at a broader detail level. We collect more detail than is required from a regulatory perspective. It supports the regulatory needs and then the loss cost and the other products. The loss costs, however, are also filed and regulated. Overall, that whole data space is audited and very strict governance guidelines. That's what really has created the underlying proprietary database for Verisk Analytics.

[[00:05:23]]
Could you share your impressions of Verisk's Core Lines Reimagined initiative and the success that has had in terms of digitising the data workflows and improving carriers' access and productivity? How is that being received by the carriers themselves?

[[00:05:49]]
Reimagine was a huge strategic and tactical undertaking when I was at Verisk and dedicated teams working on that. They went back to reimagining almost every piece of the business. A lot of focus, a lot of resources on it. I think it's been very successful. I've been reading and listening to some of the podcasts and posts that have been going on around the new product launches and how it's all been integrated. I think it's very exciting. I think it has and it will be very well-received by the carrier base.

[[00:06:58]]
Could you share a broad view of where the P&C [property and casualty] market finds itself today? Verisk has discussed how, across its business, some areas find themselves in a rate-hardening market, and others in a soft market. The company has noted that, historically and in general, organic growth has been around 7.3% in a hard market vs 6.8% in a soft market. That's somewhat of a small organic growth point differential. Could you share your impressions of the current rate environment across the P&C market? What does NWP [net written premium] growth look like? How do you think that's influencing the willingness of Verisk's clients to spend on incremental data?

[[00:07:58]]
Yes, the difference, and it does vary by product line of business, what the underlying drivers of that loss and then what goes into the rate because your loss experience is really what drives further ability to take or need to take rate and such. That said, I think there's been an ability to pivot to focus on lines that need more attention during the different market and balance out between the various lines of business. The way Verisk is structured is there are teams focused on various lines of business. They are able to direct resources as needed to meet the market need. Now because of the regulatory environment, you can't just take rate whenever there's a little blip. There is a schedule. You can't file for a rate change every three months for some lines, it has to be annual or maybe even longer tail than that. There's been a real good ability to pivot across those directives.

[[00:09:56]]
Let's talk about AI. Verisk's share price is down a little over 10% as of July 2026. I think that share performance has shown some improvement over summer 2026. In many ways, this reflects investors' uncertainty regarding the overall value dilution or disruption that AI potentially entails for Verisk's business. What are your thoughts on AI's potential impact, both from a risk and opportunity perspective?

[[00:10:43]]
I am a shareholder. I have seen the fluctuation in the market's trust, if you will, of where Verisk is going. I do believe that AI, while there is some risk, there is a lot of opportunity. I think from what I'm seeing, the changes that are being implemented and the way AI is being used to augment the strong data is really exciting. Being able to put that into workflows, I mean, actual workflows that are being used by underwriters and claims people every day as opposed to just a static, something that's in a binder, used to be, I'm ageing myself there, but just a static loss cost, I think, is really a great opportunity and is a very exciting direction that Verisk is going. I think the risk in some ways is almost an educational opportunity in that. I think if people understood more about how the data is there and is supported, there'd be less fear around how AI could disrupt and eliminate the work.

AI is not going to supersede the decades of experience, detailed data that exists. It can augment it, and it can take that data and drive significantly more value to the customer than you had with the traditional loss cost products. I've also seen how some of the underwriting products like the Coverage Verifier, which is also a contributory database, and that was one of the datasets I was responsible for. They have been able to leverage AI into making that more interactive for the customer as opposed to just something that you check, like a spot check, a snapshot in time. Again, that's something that's going to drive much more value for an organisation and then thereby creating more value itself for Verisk.

[[00:14:21]]
How could we qualify the proprietary data moat that Verisk has? The company notes that over 90% of its revenue comes from unique data or proprietary IP that is nearly impossible to replicate. How would you characterise the depth and breadth of its proprietary data moat?

As we continue to see the AI theme mature, what specific datasets or areas of Verisk's business would you call out as more easily replicable or susceptible to AI disintermediation? For instance, is underwriting an area where it's easier to scrape or acquire set data from LLMs [large language models] vs, say, loss cost? How are you thinking about that?

[[00:15:33]]
It's a very deep and broad dataset. It is decades of experience collected from a company number I have in my head is like 1,800 insurers. They vary in size from very small insurers to very large insurers. There is that whole amount of detailed data. From an organisation perspective, the other stat agents don't collect the same level of detail. There really isn't anybody else who has that deep and broad dataset. That dataset is highly managed. The quality tests and the scrutiny that has taken place on that dataset over the decades has only gotten stronger. Even eight or nine years ago, we were starting to introduce machine learning models on the quality of that, so a real forward-thinking organisation. It's not the kind of data you can scrape. There are datasets that Verisk has that are scraped. You can't scrape this experience. It doesn't exist anywhere else. You would have to go to every insurer to get that data. It's not scrapable. It doesn't happen. Sometimes I get asked, "What if companies just decide to do it themselves?"

[[00:17:55]]
You're referencing the carriers, correct?

[[00:17:57]]
Carriers, yes.

[[00:17:58]]
In recent comments from Verisk CEO Lee Shavel was asked, "What are the carriers' mentalities and AI investment strategies? Are they focused on top-line or cost-saving opportunities?" The CEO broke down what he's seeing from carriers into three categories – (1), the first priority for the insurers is overwhelmingly focused on efficiency and cost-saving opportunities, (2), portfolio and risk selection applications and (3), customer interactions themselves. One of the sentiments from the investment community around AI stems from the idea of, "What happens if the P&C customers and insurance carriers themselves choose to keep that data internally, as opposed to sharing it with Verisk, and run their own analysis upon it?" Given carriers' priorities around efficiency and generating cost-saving opportunities, what do you make of this argument? Do you think there's validity to such a scenario taking place, at least in the medium-to-long term?

[[00:19:05]]
I have heard that for years. The response is if a company… First of all, you have to have a lot of data to do that. If a company decided to do all of that on their own data, then their experience really is only of their own data. The value from the data moat is it's data across the industry, across different insurer sizes. You get a truer picture of the industry than you would if company, ABC decided to keep their data and do their own analytics. It's like making a cake and you keep all the ingredients separate. It doesn't come out the same way. That's a true risk for an individual carrier. You also then have the other position that the small companies can't do it because they just don't have either the depth of data from their own database, especially as you have new insurtech coming into the space, new insurance companies popping up, they really need data from outside. They just don't have the data. As I said, that core lines data is not scrapable. It doesn't exist anywhere. It's homogenised, which is actually what you want for a loss cost. You don't want to base your loss cost on your own data and your own strategy because then you miss evolving trends.

[[00:21:26]]
From a cost-benefit analysis lens, and assuming the continued maturation AI capabilities, when we think about a large carrier – as opposed to a mid-sized or smaller carrier which, as you mentioned, wouldn't really have the incentive to do this analysis themselves – if cost savings and increased efficiencies are the top priorities for a large carrier, how are they thinking about the cost-benefit analysis of disintermediating from Verisk and running said data analysis themselves?

Is there a point at which you could see the carrier come to believe that running their own analysis is more cost-advantageous over continuing to choose Verisk's platform at a higher cost? Or does it make more sense to pay Verisk a premium, given its proprietary data moat and the characteristics that you've assigned to its business in our discussion so far?

[[00:22:49]]
From my perspective, an insurance carrier is in the business of underwriting and settling claims and market strategy. I would say it's better for them to focus their resources on those core, I keep using the word core, those areas of the business that are going to drive their revenue. It takes a lot of resources to collect and manage data and to go through all of the data quality. I think even for large companies, that's resources that could better be spent on driving business results. Early in my career, I worked for a company, an insurer. Even back then, and this is in the '80s and '90s, we always talked about, "Why are we sending data? Why don't we do this ourselves?" Ultimately, it came to… It was expensive. It just made more sense to have somebody else who's in the business of data, manage the data and let us drive results through underwriting and product development and claims handling. When I was in the data area at this insurance company, and we were a cost centre. We were not a business centre. It's thinking about it that way, where for some companies, they do invest a lot in data because they do their own things with it. There is a portion that is just… It makes sense to have somebody who can manage this data at scale.

[[00:25:19]]
Let's talk about some other credible or competitive threats. As we discussed, Verisk characterises over 90% of its revenue as coming from unique or proprietary IP that is nearly impossible to replicate. Conversely, some investors think that roughly 30% of Verisk's business – specifically in life, auto and international analytics – is potentially disruptible by new AI-native competitors who could potentially replicate the data points or sets that Verisk has.

So far, the company's management team has claimed that no credible AI-native competitors have emerged yet, and that, rather interestingly, these AI-natives are actually seeking Verisk as a distribution partner vs directly competing with it, given the company's extensive proprietary data moat. How would you assess the risk credibility that AI-native competitors pose to Verisk's business as we look out over the next 24-36 months?

[[00:26:40]]
The automobile business, I think, is less disruptible because it is experience-based that has to come from the companies. For the life and international, I'm less familiar, except I do know that some of that stuff was scraped. You can scrape some of that data. Again, I think it does come down to economies of scale. If you are an AI company, why wouldn't you partner with someone who's already done the heavy lifting on the data piece and you're focusing on the analytics? I think that's why the partnership sounds so interesting and reading up on some of them, it really is business-changing.

[[00:27:56]]
Verisk talks about perceiving AI as being, in many ways, extremely effective in lowering the barrier to product creation. That seems to be where many of the AI-natives have positioned themselves – in workflow enhancements such as document digestion and agentic process automation, rather than directly trying to replicate Verisk's deep proprietary data moat. To this end, the company has been adamant that it controls the distribution and data channel and that these remain binding constraints that insulate its business from AI disruption risks. Do you consider yourself aligned with that view?

[[00:28:56]]
I do. From what I saw before and what I see now, they sound like really exciting partnerships. Verisk has been very clear in stating that the data is fully protected and trusted and governed. I think that's the million-dollar value that a company like Verisk brings to a partnership.

[[00:29:45]]
You brought up partnerships, and we've seen Verisk announce partnerships with several frontier model providers, the most prominent of those being Anthropic. To reiterate some of the distinctions you mentioned, Verisk has said that it controls what data is sent through these MCP [model context protocol] connectors and more importantly, that these frontier models cannot train or receive the company's proprietary data. In theory, this means that its complex analytical capabilities would not be replicable by said frontier models and thus, would require additional advanced tooling, which only Verisk can provide, given that it sits on decades of data calibration and resources. Could you share your thoughts on the recent frontier model partnerships? Moreover, do you see these partnerships as more additive or subtractive to the business's value proposition going forwards?

[[00:31:09]]
I definitely think they're more additive. I think being able to partner agnostically will drive more value to Verisk. As the frontier models leverage the data, they will drive more value and more demand, which will then cycle back. I definitely see it as additive. I don't see it as cannibalising the Verisk analytic capabilities in any way.

[[00:32:03]]
Verisk's management has talked about how the insurance industry, an industry punctuated by a risk-averse nature, is eager around technology adoption, but very deliberate in adopting these new AI technologies. Going forwards, what role do you see the LLMs and frontier model providers playing within the broader context of the insurance industry? Should we consider this to be an incremental distribution channel for Verisk? Or conversely, in 24-36 months, for example, could we see these frontier model providers morphing into significant competitors to Verisk's business?

[[00:33:15]]
I don't know about morphing into… There's always the possibility that these providers will morph into competitors in certain spaces. Verisk does offer analytic products as well as the core lines products. That said, I do think that Verisk probably has a lot of resources internally building their own AI capabilities to also present to the market. I do think I saw in an investor report that the amount of tech spend has really jumped. That's because all of this AI work. Just seeing on LinkedIn this morning, they're hiring specific AI roles, AI governance. They're not sitting back just figuring to partner. They continue to build their own AI knowledge capabilities. I think it will be a strong partnership. I think you're going to see both Verisk coming out with their own and the providers ultimately becoming a competitor, but against products that Verisk will have developed.

[[00:35:26]]
It's important to remember that Verisk isn't operating at a standstill. The company is innovating. For instance, it has come out with tools such as XactAI, its AI-powered suite of tools integrated into its Xactware claims ecosystem. To that point, how effective do you think the organisation has been in terms of embedding AI across multiple product suites?

[[00:36:04]]
That started I think even pre-COVID… At the beginning, before a lot of companies, before it was the buzz that it is, it was buzzing at Verisk. They were looking for ideas and really coming through for test cases to open it up to say what would our business look like with AI, where can we put AI in and where can we drive more value. They were doing this in parallel with Reimagine, and that was also part of the Reimagine. Let's Reimagine workflows and let's think about what AI is and what it could be.

[[00:37:29]]
Verisk has often claimed that AI implementation on the client side will ultimately lead to increased data consumption by the carriers with Verisk. Are you in alignment with that viewpoint? Are there any anecdotes from your time at Verisk that you can lean on to support this argument?

[[00:38:19]]
Definitely AI will lead to more data consumption because by putting it into the workflow, it will be very transactional with the business. You check something at the beginning of the day, and that's it. It's going to be highly interactive. That's where the consumption is going to be. As it gets embedded into the business workflow of the insurance company, they'll make improvements and they'll see efficiencies that will drive new product innovation for them because, again, a more efficient workforce can do better, different things. Then again, that will drive back to more data consumption. As far as anecdotes from my time, just that they sought test and use cases from the employees, the employees who were working with customers and knew their pain points. I think even five years ago, there was the drive to develop and bring something into the workforce that would be new and exciting.

[[00:40:19]]
Verisk has expressed a belief that as AI-enabled Verisk tools gain wider adoption, that token costs will remain manageable, in large part due to the fact that these are borne by the customer. More specifically, the compute workload of these AI-enabled tools is performed on the customer side, subsequently meaning that Verisk does not bear the ultimate accrued costs of incremental usage. Would you largely concur with this talking point put forwards by Verisk’s management team?

[[00:41:07]]
Yes, I absolutely would agree. Even for some of the products several years ago that there was no AI embedded in them. The transactional costs are really borne by the customer and how much they interact with the data and the product suite. I think from an AI perspective, if it's embedded in the workforce, the work stream, it's going to be the customer who is kicking it off and who's interacting. Yes, that's where the brunt of the cost is going to be. The workforce is going to be more efficient and be able to drive better insights, which would then hopefully lower operational costs on the customer side. I think I don't want to give the impression that there's going to be this huge cost that's passed on to the customer and the customers left holding this big invoice. I think it will be offset by the efficiency that it's going to drive and the better decisions and better decisions mean better risk, and that should drive better results for the company.

[[00:43:02]]
I think Verisk has talked about around a 5-10% productivity uplift, whether in claims or underwriting, through the usage of AI. That said, I imagine this is on regulators' minds, particularly in an industry such as this. Is there anything that you think investors should be aware of in terms of the regulatory landscape and things that could change there over the next five years? Do you see anything that could impede the adoption of these AI tools within the industry, or conversely, accelerate their adoption?

[[00:43:44]]
The regulatory environment does not move quickly. You are talking about an industry that is very risk adverse and the regulatory community is slow for adoption. I think at its core, the pricing of the insurance product is highly regulated. I don't see that changing. Where I do see… I know the regulatory has been looking at for years studying even predictive models and how much they should be regulated and what can be done and what can't be done. That said, I think there's a fair amount of time ahead before anything is going to change.

[[00:45:08]]
Whether positive or negative.

[[00:45:10]]
Whether positive or negative.

[[00:45:13]]
I think you mentioned having seen that tech spend is on an upwards trajectory. To this end, Verisk does indeed face some costs in the process of actually embedding AI into its own product. However, management expects to reflect this cost in future monetisation efforts. How cost-intensive do you expect the embedding of AI tools across existing products to be, and moreover, do you see said costs being outweighed by future monetisation efforts or increased data consumption by customers?

[[00:46:08]]
I'll start with the last bit. I don't think costs are going to be outweighed by future monetisation. I think there's going to be an initial hit as you gear up. Verisk is really good at managing the operational cost of its business. I think it will drive new business to Verisk, and I think they will be able to monetise differently, but make it so that it is, of course, of value to Verisk to undertake this.

[[00:47:20]]
How cost-intensive do you expect the embedding of AI tools to be across existing products? I think you alluded to this earlier when you said it's on an upwards trajectory now and could potentially sustain that going forwards as Verisk looks to innovate and stay ahead of the growth.

[[00:47:39]]
Yes. There is a significant cost upfront. I think leveraging partners is a really strong idea for the lessons learned and maybe knowing where they want to put their focus in which tools and which segments of the business and which enablers they want to build. I think ultimately, Verisk will be fine. They will figure out the sweet point, monetise and be able to drive value for themselves while driving value back to their business, their customers.

[[00:48:31]]
In terms of improving underwriting or claims efficiency, how much of an incremental improvement do you think AI can lend to carriers?

[[00:48:45]]
From what I'm reading just in the underwriting portion, putting AI right upfront where you're doing the quoting or the binding of a policy, being able to go out and get that knowledge right upfront as opposed to the legacy aspect of underwriting. It's going to be much faster. it's going to be real time. That is a huge cost savings for a company. I think there's a great opportunity in the embedding of this in the underwriting products. They're doing it at point of sale. They're doing it even before there's now the capability of letting a customer know when their customers are shopping before their policy period comes up, because retention is an issue in the industry, depending on the line. They're finding that because of cost, customers are shopping earlier and more often than they necessarily did before. There's now the analytics available to customers that say, "[A customer] is starting to shop around for a policy. Is this a customer you want to keep? Are you happy to have her shop around?" I think that's a really strong use case for integrating AI into the business process.

[[00:50:41]]
When Verisk talks about improving underwriting efficiency by 5-10%, or at least hints at that, do you think that's conservative in terms of the overall efficiency improvement that we could actually see going forwards?

[[00:50:57]]
Yes. If nothing else, I think Verisk is measuring conservatively, because it's also depending on the size of the customer company and what their pain point is in the underwriting cycle, the ability to focus and drive a better response in real time, I think it's going to be more than 10%.

[[00:51:38]]
What do you think investors need to see in order to instill confidence that AI isn't a disruptor or disruptive element to Verisk's business, but rather a powerful tool to help drive accelerated top-line growth or an improved EBITDA margin profile?

[[00:52:04]]
I think education like this is very important because you really have to understand the foundation that Verisk was built on and how that foundation is irreplaceable, I'm going to say, and maybe that's a strong word. It really is such a strong foundation. It is so deep and so wide and so trusted and protected and regulated that the underlying data is not going to be superseded by AI. AI is going to use the data and improve the business cycle. I think it is a matter of having these conversations and that if the investor community really understood how Verisk became what they became and where the data is and where the data sits, and how that is the fuel for everything. I mean we used to say data was the oil and we were the factory processing this and the amount of attention to detail that goes into the underlying bedrock of Verisk. I see AI as a very exciting enabler. I don't see it as a replacement.

[[00:54:16]]
Early in 2026, I think Verisk's management team upgraded its pricing contribution target to 450bps, up from 400bps. That may have changed, but nonetheless, there seems to be a strong pricing motion. That's built on the back of product innovation being the primary driver. What do you think drives the price algorithm at Verisk? When we balance the elements that have come up in our discussion – innovation vs cost-conscientiousness by the customers or carriers themselves – do you think the company can continue to take price in this market?

[[00:55:17]]
That's a big question. I think Verisk can continue to take price as or when they demonstrate the significant value that can be delivered through these new products, these new offerings, these new enablers. Price is always a discussion that we had with the customers. Why am I paying so much for this loss cost? People are willing to pay for value. I think especially with these workflow enhancements, that are being put out there, I think customers are going to see the value and priced appropriately, they're going to pay for it because they're going to get more business value than the cost they're paying. I'm highly optimistic.

[[00:56:54]]
Let's talk about Verisk's top and bottom lines. We've seen quite strong EBITDA margin improvements over the last several quarters within the business. Those margins are around 56% today, largely driven by global talent optimisation and partially offset by some increased investment in AI and technology. How would you weigh up those two factors? What are your thoughts on the company's ability to control costs internally within the organisation through corporate controls, talent optimisation and so on vs the increased investment that's required to remain at the forefront of these AI technologies? Presumably, when these AI technologies are successfully infused across the organisation, once that upfront investment has been made, they will yield dividends in terms of profitability. What are your thoughts on that?

[[00:58:03]]
I was really excited about the global talent optimisation. I had a team that was located in Poland. We found significant value in that optimisation, where we could manage costs better than we could without staff that might have been metropolitan-based, and the calibre of talent was unbelievable. I'm very proud of the work that we did. It was actually one of my directors who went out and built the team, the first global team. We were a big proponent of it years ago, and that really did drive a lot of operational efficiencies that we had.

[[00:59:19]]
Historically, Verisk has grown in the mid-single-digit range. What's your top-line organic growth outlook for this business for the rest of 2026 and beyond? Should we think of this as a business that's growing between 3% and 6%? Could it reach the high-single-digit range, perhaps 7-9%, due to the themes and potential tailwinds of those themes that we've been discussing?

[[01:00:26]]
I think the potential for larger growth is definitely there. Again, it's about perceived value. If the customers see value in what's being delivered through these new products, they're going to be willing to pay for them. You get the buyers angst when they don't see what they're paying for. I think by improving workflow and shortening underwriting cycles and shortening claim cycle, that's measurable. It's measurable from an insurance perspective. What they can measure, they can see the value and then do the benefit analysis and come out that, "Yes, okay, we're paying more, but look at how much more we're getting." I'm confident.

[[01:01:39]]
Could you share your impressions of Verisk CEO Lee Shavel and the broader management team at the company?

[[01:01:54]]
I worked with all of them. Some of the smartest people I have ever worked with. I think it's a really strong team. I was a consultant before I joined Verisk. I've been in a lot of other companies. I've seen the calibre of different leadership teams. I think Verisk does have an extremely strong and extremely smart and business savvy management team. If anyone can deliver, I think this is the team for it.

[[01:03:11]]
We will end the Interview here. Let me close by saying thank you for not only taking the time to speak with us but for the wonderful insights. Thank you, clients, for choosing Third Bridge. We hope you enjoy today's content and a big thank you to all the team members who helped put this discussion together today. Bye.

[[01:03:28]]
Okay.

Transcription ends at 01:03:31 of the recorded material