Which Insurance Tech Solutions Are The Most Insulated From AI?

As I noted in recent posts, I had been working on a big piece about how AI could impact insurers including threats to data providers, brokers, D&O, the foolish carriers writing data center risk, and so on.

That piece is now in the trash bin as current events moved too quickly to address them all preemptively. Some (mostly on the underwriting side) still have not impacted the market and I will hopefully get to those before they become old news too.

However, first I wanted to address some of the fears that are already in the market. I will save insurance brokers for later and start with the information service providers.

This refers to a broad batch of companies such as:
– data solutions products from Verisk, CoreLogic, Cambridge Mobile, and LexisNexis which are primarily underwriting focused
– agency management systems like Vertafore and Applied (mostly owned by PE or VC firms)
– policy admin systems like Guidewire, Duck Creek, etc.
– industry data aggregators like the former SNL, now part of S&P Global and, to some extent Verisk again and CCC
– the online lead generators such as EverQuote and MediaAlpha

For the purposes of this discussion, I’m going to leave out the comparative raters as I think they are more germane to the broker challenges.

So, you can see it’s a fairly broad range of services involved, yet they all have been targeted for extinction if you believe some of the more sensational headlines.

I’ll try to briefly assess the risks to each type of service but I think, first, we need to set some parameters around what features would make these businesses more or less at risk of being disrupted.

Barriers To Entry

Let’s first maybe start with a model that would be easy to disrupt. Anything based on publicly available data has no moat, so if you’re selling repackaged economic data, you’re probably in big trouble.

Similarly, if you’re aggregating and repacking a service (e.g. a travel agency), that’s going to be easy to disrupt.

If you built an online advertising or web design agency, it’s probably time to think about plan B. A business model based on being really good at generating low cost online leads or high rates of click through is ripe for AI competition.

However, if you have propriety data, or data that is a result of some proprietary process, then you are in better shape. AI will have a hard time accessing that data set.

What else is helpful? Processes around the data. If you take data and bundle it with an underwriting or claims solution, that is more difficult for AI to imitate. It might be able to make a poor replica, but it’s likely going to lack the specialized knowledge required to optimize the result.

Speaking of specialized knowledge, this is one of the biggest barriers, especially if it requires a lot of upfront learning and doesn’t have a giant end market.

AI will be motivated to gain specialized knowledge for trading stocks. I can’t imagine the folks at Anthropic are dying to figure out how to lower LAE costs. Being boring is a win!

Finally, I’ll add good old inertia. We all know a big reason retention is so high across the industry, even though you can “save 15% in 15 minutes” is people are lazy and don’t want to think about insurance. Inertia is also your friend!

Relatedly, I’d add that there are indirect costs to switching. Even if AI has a better solution, is it worth the time and effort to train all your people on it? After all, Google Sheets is free yet most businesses still pay for Excel. My guess is a lot of AI offerings will be more like Sheets.

Build In House First

Beyond barriers to entry, what other options does a software business have to defend themselves from an AI disrupter? How about build it in house yourself? Funny enough, all the AI businesses are happy to license you access to their models!

In fact, I would suggest AI is more of an opportunity than a threat at many of these companies. If you can use AI to improve your existing product, then you can actually raise your price and your customers will gladly accept it.

Consulting prices didn’t go down when Excel came along. They went up because consultants could provide better answers faster and were thus more valuable. Why would it be any different this time?

Of course, many existing businesses will struggle to adapt and/or not execute well, but there will be plenty who do see better growth and margins as a result of AI. Let’s not forget, the company who figured out how to apply the internet to insurance was Progressive, not a newco.

This Movie Is A Remake

This is not the first time people who don’t know insurance have predicted large disruptions to insurance. Most recently, there was the startup crowd which are still trying to figure out how to make an underwriting profit.

Before that, there were the intermittent efforts by Amazon and Google to launch insurance pricing sites which would lead to immediate selloffs in the large insurance stocks only to be forgotten about within a few months when they inevitably went nowhere.

Even beyond insurance, the internet bubble led to similar panics. Remember when every retailer and restaurant was going to 0 because we would never visit brick and mortar stores anymore? Some retailers did go bankrupt and others had to adjust their models, but we still have a thriving retail sector today.

And so it will be for most other industries. There were clear losers in places like media and there will be losers this time. Software, in general, seems like a good place to look for those losers, but it doesn’t mean it will be widespread or that it will extend to niche areas like insurance.

It is human nature to overestimate threats – “shoot first, ask questions later”. We are in the overestimation phase at the moment.

Eventually, cooler heads will prevail and the market will be more discerning about which companies are at risk and which will figure it out. What follows is my first draft of trying to figure that out.

Data Solutions

This is Verisk, CoreLogic, etc. These businesses collect industry data (loss runs, claims handling, miles driven, home features, etc.) and analyze them to produce products that improve underwriting or claims outcomes.

It’s very hard to figure out how AI providers will get access to this data. Without it, they have nothing to train on and can’t build a competing product.

Is it possible the large insurers would stop providing data to Verisk and give it to Anthropic instead in hopes of getting a better product? I guess they could, but I’m not sure why it would be better.

The reason some of these products are predictive is they have a) lengthy historic data and b) diverse data, meaning much of the industry participates. If you can only give Anthropic your last few years of results, and it’s only for 20% of the market share in a line, their analysis isn’t going to be as robust.

I suspect inertia will be a big benefit here as underwriters are comfortable with the tools they know and may hesitate to change. I also don’t think the AI companies are going to be super excited about the market to replicate CLUE reports.

Given the data solutions companies already have the data and products though, they should be highly motivated to license AI to make a better in house version of what they have. If they can demonstrate it lowers loss ratios, companies will accept price increases on that new and improved product.

Thus, I think this group probably has more opportunity to gain than lose from AI.

Agency Management Systems

These are the software vendors that help insurance agencies manage their business from tracking leads to getting quotes to billing and renewals to compliance and so on. These systems have increased in capability and complexity over time and there is a constant flow of new entrants promising to one up the best in class provider.

These systems don’t require any special access to data. The agent is going to provide the same info on their clients to any AMS vendor they use.

There are somewhat significant switching costs as agency owners will be reluctant to train CSRs on a new system if there isn’t material benefit.

There is very little in these systems that is truly proprietary. When one provider brings a new feature, other competitors typically try to match it. Vendors compete on ease of use, compatibility, features, and obviously cost.

There is a history of new competitors entering the space with more modern offerings and gaining share from legacy providers who can’t offer the same functionality due to legacy issues.

In other words, this sounds like a space ripe for AI disruption. However, that doesn’t mean one of the AI firms will decide to build an AMS. More likely, a startup (or two or ten or even a hundred) will try to use AI to improve on existing offerings.

There are plenty of stories about startups seeking to automate back office functions, “read” documents, verify the accuracy of customer information, etc. Surely, one of them will find success.

So, I’d definitely be concerned if I were a PE that owned one of the large AMS firms, but I think the threat is no different than the past – some startup will use the newest technology to build a better mousetrap than the prior cutting edge version.

Policy Admin Systems

Most people think of this as Guidewire but there are other private competitors like Duck Creek, OneShield, etc. These companies are the IT backbone for underwriters handling submissions, quoting, billing, claims, data warehousing, etc.

These systems are much more integral to underwriter operations than an AMS. They can be nine figure spends at larger companies.

Given how core they are to operations, it is extremely difficult to imagine an underwriter switching to a new, unproven system. The switching costs are very high, not just financially, but in terms of pain from a poorly executed new system.

There are plenty of issues with Guidewire that insurers would very much like to see improved, but they’re not enough to risk a switch to something untested.

If a new option emerges here, it will likely have to target small MGAs who can’t afford Guidewire. That is obviously a growing market but there are a number of existing solutions there already focus on that space.

Potentially, AI could make it more viable for a tech focused startup to build its own PAS rather than buy something off the shelf. This would be a concern for those smaller, emerging PAS vendors but not for the Guidewires or Duck Creeks.

Data Aggregators

The most obvious name I have in mind here is S&P Global which is a wonderful service at making stat filings and other industry data easy to use. However, there is no moat around the data here. I could probably build an app in Claude today to compile every stat filing.

It is not hard to imagine AI building an imitation of the old SNL platform and charging far less for it, even while offering as much or more functionality. The same could be said for other data vendors who compile public information from highway driving info or workers comp claims among other things.

These are the models I find most at risk.

Online Lead Gen

These are companies like EverQuote and MediaAlpha who specialize in generating leads for auto insurers. Their stocks, like other advertising centric business models, have been hit hard of late – and for good reason.

The value add in these models is largely optimizing search engine and social media results through constant experimentation. This is something AI should be especially good at. AI can also probably do a better job of quantifying the quality of lead resulting in more efficient pricing for carriers.

The real test here will be whether the companies can find a way to integrate AI themselves before someone else comes up with a better solution. I don’t have a strong view on this yet, but if they don’t find a way to use AI themselves, they could be in a lot of trouble.

It’s Still Very Early

It goes without saying that all of these thoughts are very speculative. The market’s job is to handicap probabilities. Nobody knows with high conviction where things are headed. However, the weighted average of the increased likelihood of negative scenarios means the stocks should be lower to some degree.

My guess is we will look back in a few years and find some companies adapted well and this was a great time to buy while others put their head in the sand and got disintermediated.

There are a lot of possibilities I didn’t cover, either because I didn’t think of them yet, I thought of them but forgot to include them(!), or I was wary of making this piece any longer.

I am sure I will revisit this topic down the road once we have some more data points. I will try to get to the broker side of things next time.