The logic went: if consumers can quote and compare inside ChatGPT, why use a broker at all? The threat was disintermediation: folks would simply cut out the broker as the middleman when buying insurance.

For Meyer Shields, managing director and equity research analyst at Keefe, Bruyette & Woods, this extrapolation seemed a bit dramatic.

“This is not the first time that we’ve seen some sort of ’apocalyptic’ scenario for the insurance brokers. I’ve been in the industry a little more than 30 years, and it comes up a lot,” Shields says, saying it’s the same argument that was made at the dawn of the internet.

The result? Three decades on, there’s been some success with direct-to-consumer car insurance online, but there are still basically just three main players in that market: Progressive, GEICO, and Allstate.

“Fundamentally it does not look like people want to buy even the very simple products using technology as opposed to an agent … I look at homeowners penetration in direct-to-consumer. It’s probably less than 10% — maybe 7% or 8%. And next to auto, that’s probably the second-most homogeneous insurance product,” Shields says.

So if the AI drama is overblown, why has the impact on market prices endured?

As of the closing bell on April 17, all six broker stocks mentioned above still traded below their closing price on February 6 (the last trading day before GPTgate), despite the S&P 500 adding a healthy 2.8% over the same period.

What gives? Does this industry have a new risk profile due to a secular change in the way of doing business, or is this just another example of Wall Street myopia?

While the ChatGPT news sparked disintermediation fears for personal lines, that threat looks far less immediate on the commercial side.

“I do think it’s going to affect top-of-the-funnel kinds of markets more than commercialized markets. So think personal lines, homeowners, and personal auto — more than complex advisory risks and things like that,” says Farah Ismail, head of commercial lines for ICT (Insurance Consulting and Technology) at WTW.

Brokers profit from information asymmetry, and AI levels that uneven playing field for simple products. If you’re factoring in AI as a major new force threatening to change business as usual, look for personal lines to be tested at the vanguard.

Conversely, “specialty, commercial, and multi-line risks need bespoke coverage. There’s more underwriter judgment, there’s more negotiated terms … and from a broker revenue standpoint, the bulk of that revenue sits on the commercial, specialty, multi-line side,” Ismail says.

Tracy Dolin-Benguigui, director and senior research analyst covering life and non-life insurance at Wolfe Research, hammers this point home with an example.

At a Gallagher (AJG) investor day in March, AJG “basically said they had identified $500 million of premiums that they place that are either personal lines or micro-commercial, and then they took a deep dive and really only identified $200 million of those premiums as more susceptible to AI risk,” Dolin-Benguigui says.

“That represents something like 1% of their global revenue” after accounting for commissions, Dolin-Benguigui notes. “So it’s pretty contained.”

Raphaël Vullierme, co-founder of WaniWani, the AI distribution infrastructure that powered the Tuio/ChatGPT insurance app integration, agrees that the risks to brokers are contained — for now.

In the short-term, he admits that complex and commercial risks are essentially safe from disruption. Risk managers want to go through real people, and more commoditized personal lines make for easier pickings.

Look beyond the next few years, however, and Vullierme — who himself co-founded Luko, which grew to become France’s largest online home insurer — thinks things could change quickly.

“There’s nothing really complex in matching risk capacity and demand that AI in 10 years won’t be able to do,” Vullierme says.

The internet helped commoditize the sale of goods, Vullierme says, but two frictions heavily insulated insurance brokers: buyers needing to provide complex personal inputs, and the difficulty of comparing bespoke products.

He sees AI dissolving both.

Specifically, Vullierme cites a catalyst called MCP (Model Context Protocol), a standardized protocol allowing AI “agents” — as in, software programs that act autonomously — to interact not only with external data but also with other AI agents.

Introduced in November 2024, MCP was developed by Anthropic, the firm behind the Claude AI chatbot.

“It’s your AI as a buyer that’s going to speak to the AI of the seller and say, ’This is a business doing $10 million revenue, specializing in distributing pet food in Iowa.’” The technology now exists for your AI agent to communicate with the AI agent of an insurer to derive a customized commercial quote for your approval.

“This is going to create a lot of disruption in the market because two AIs can work together to build a custom offer. This was completely impossible two years ago without a human in the middle working for both sides. Now it’s possible for every kind of risk and every kind of business line,” Vullierme says.

While the technology might be ready, the U.S. regulatory landscape is not — and that’s a meaningful speed bump for AI-driven insurance distribution.

Spain’s Tuio operates under the luxury of a single European regulatory framework, while insurers in the U.S. navigate varying regulatory environments governing how policies can be marketed, quoted, and sold.

One example: Colorado’s SB 21-169,⁴ signed into law in 2021 and among the first AI-specific insurance regulations in the country, requires insurers to demonstrate that their AI-driven systems don’t unfairly discriminate against consumers. The law places the compliance burden squarely on carriers and any technology partners they use — meaning an AI app offering quotes inside ChatGPT would need to clear a higher bar in Colorado than in most other states.

Nationally, the NAIC’s Big Data and Artificial Intelligence Working Group⁵ is developing model frameworks, but adoption across states has been uneven, creating a regulatory patchwork that slows any national rollout.

None of this means AI-driven distribution won’t arrive in the U.S. But it does suggest that the timeline for the kind of seamless, in-app insurance purchasing that panicked investors on February 9 is longer — perhaps significantly longer — than the market’s one-day reaction implied.

The groundbreaking idea behind Tuio’s integration into ChatGPT — and a major reason insurance stocks took a beating on February 9 — was that customers could get personalized quotes within ChatGPT by answering questions without ever leaving the chatbot. But there’s a nontrivial part of that story the market seems to have glossed over.

“In reality, the app suggestion part from OpenAI is not working very well at the moment,” Vullierme said in a March 23 interview.

The idealized user flow has ChatGPT suggesting the relevant app for the user to utilize within the ChatGPT environment.

“But for now, the suggestion part of the app is not working very well, so the user needs to install or say ’@’ the name of the app to use it.”

Not only is the industry-killing AI chatbot integration not the frictionless disruptor people feared, there are also plenty of ways that AI is being deployed today that stand to benefit both brokers and carriers.

Shields says that both brokers and underwriters have expenses that should decline over time due to efficiencies brought on by AI and tech-enabled efficiencies. That’s a win-win-win for brokers, carriers, and consumers in theory, because in a competitive industry some of those lower costs will be evidenced in the lowered premiums.

Shields gives a simple example of how AI can shave costs for brokers: “If somebody submits an application and they’ve written in the address field using words like twenty-six instead of the number 2 6, that can take some time to digest, and we’ll soon, if we’re not already there, get to the point where that’s very easy to digest. That sort of small inconvenience adds up to a real cost to brokers that will go away.”

In a fragmented marketplace, Shields thinks brokers who invest in technologies that reduce the time to get customers a quote will see benefits accrue.

“The same way Progressive and GEICO have lowered overall industry costs because they’ve become bigger and their costs are lower, I think the big brokers will have that benefit as well,” Shields says.

For carriers, “the time between getting a submission from their broker and then being able to turn around an actual quote — that time has reduced quite substantially,” Ismail says.

Ismail also sees retention rates benefiting from automation “because you already did your due diligence the first time around. So you might be able to create much more robust guidelines on renewal to say, ’Look, as long as your exposure hasn’t changed dramatically, this is where we are on pricing.’”

Thus far, the ChatGPT disintermediation story is only a story. And while stories play an important role in short-term Wall Street gyrations, the early days of the ChatGPT-insurance app experiment have been anything but a slam dunk.

Shields, who strongly believes the February sell-off was overdone, succinctly sums up how he thinks about the “AI will kneecap brokers” narrative: “I think the upside to the brokers from making this a faster process far exceeds the few people likely to buy insurance without an agent.”

If AI makes policy shopping meaningfully easier over time, is it fair to assume that actuarial assumptions around lapse modeling and customer lifetime value will need revisiting?

Shields thinks so, but with a caveat. “It will impact retention, but I do think it’ll be gradual. All of these things manifest themselves much more slowly than modeling them on a spreadsheet might anticipate.”

And Ismail, while she’s seen an increase in retention in the short term, acknowledges it’s a double-edged sword if AI, for example, makes the submission ingestion process so seamless that buyers can “send [their] account to 12 different brokers at every renewal cycle and potentially make changes to [their] actual policy.”

For Vullierme, who has the benefit of seeing the cutting-edge AI use cases before the general public even knows they exist, there’s one nascent practice that could challenge traditional rules of thumb around quote-to-bind ratios.

“We have seen businesses implement procurement [AI] agents, so at the renewal date, there’s automatically a procurement agent that browses for new quotes without the business having to do anything,” Vullierme says.

He says AI agents may already be ringing call centers undetected, creating a higher volume of quotes, and dinging the quote-to-bind ratio or increasing churn.

But the changes aren’t all bad for industry mainstays, Vullierme says.

In commercial lines, he’s seeing people use ChatGPT or Copilot connected to their work environments. This means these programs are plugged into all sorts of other software — document management services, accounting software, payroll programs, contract management services — the whole kit and caboodle.

This allows businesses to answer quote questions far more easily because “AI is prefilling all the answers on behalf of users and the user just has to validate,” Vullierme says.

The ease of use allows insurers to ask new, more granular underwriting questions, creating new variables and new opportunities for pricing and underwriting, he says. “We think this will increase even further the power and importance of the pricing team.”

Of course, the degree to which organizations are embedding AI tools within day-to-day operations also presents new, potentially very large, risks that come with overreliance on any one software or process.

Ismail also sees this as an area of increasing relevance and, from an actuarial perspective, thinks there are some pretty important gray areas that need to be clarified.

“As carriers are rolling out some of these coverages, they have to really rethink and look at the terms and conditions and say, ’Am I actually covering if you use Copilot incorrectly, and should I be covering that? Or does that sit on a different policy?’”

“That is very early on in discussions. I think a lot of people are now concerned about that, but I don’t think there’s a hard and fast rule to say, ’Look, your misuse of an AI tool should sit with this policy versus that policy,’” Ismail says.

And more than just a renewed focus on what terms and conditions stipulate, AI may open up entirely new lines of business.

The AI mania is driving an infrastructure spending boom, with “hyperscalers” rushing to build out data centers, some of which can cost tens of billions of dollars to build.

This is driving a surge in specialized consulting. Marsh recently launched a specialized advisory group to advise clients on digital infrastructure buildouts.

Insuring data centers at this scale means answering novel questions about how GPUs, or graphics processing units, will be depreciated in the case of a loss, how insurers can spread concentration risk across these large, unique projects, and how to structure bespoke policies for lenders in an ecosystem where financing often occurs off balance sheet via private credit. (The ability of specialized circuits like GPUs to process enormous amounts of data makes them a critical component in data center buildouts.)⁶

Insuring obscure financing deals that ultimately fund the plumbing of AI is one thing, but on the software end, the newest generation of frontier AI models could pose even more abstract challenges to traditional loss assumptions on the cyber side.

For example, Anthropic’s newest model, Mythos, reportedly has such powerful autonomous coding capabilities that its release, as of April, has been withheld from the public due to fears that it could exploit vulnerabilities faster than companies can fix them.

Mozilla said⁷ it used the program to find and fix some 271 bugs in its Firefox browser; banks are rushing to gain access to the program; global regulators⁸ are examining the risks it could pose to the financial system; and Anthropic’s CEO went to the White House⁹ to discuss Mythos and cybersecurity concerns the model may bring to light.

Even Vullierme, a gung-ho AI bull, doesn’t think brokers will go extinct tomorrow. Behavioral patterns are too entrenched, corporate risk managers too conservative, trust in new technologies too tenuous, regulators too rigid, for an insurance app in a chatbot — and one whose integration has been wonky thus far — to upend an entire industry.

For brokers, “the assets are in the elevators; they’re people,” Dolin-Benguigui says. And, for commercial lines especially, it’s still very much a people business.

She also sees another advantage for brokers in the current market: the pricing cycle. She believes the insurance industry is entering a soft market — one that she thinks may last for a while — and with insurers competing more fiercely, they’ll need distribution from brokers.

So, how does the job of the broker look five years from now?

For Shields, even given the pace of innovation in AI, he thinks the industry will simply be incrementally more efficient — not fundamentally different.

Longer term, however, it could be a very different story, although it will follow a familiar theme: consolidation. “The difference between AI-enabled and non-AI-enabled companies in 20 years is going to be so dramatic that smaller companies will really struggle,” he says.

What about actuaries? What does all this mean for them?

For Dolin-Benguigui, she’s not overly sanguine about the job security. She advised her 18-year-old niece, who’s good at math, to study to become an actuary. But given the prominence of AI, her niece was afraid there wouldn’t be a role for her.

“So I think actuaries should be afraid for their jobs,” Dolin-Benguigui says. “I would advise actuaries to try to hone their skills and think of ways that they could add value in a world where the more scalable, repeatable kinds of tasks could be replicated by AI.”

For Ismail, however, she sees all this as invigorating for the profession.

Take reserving actuaries: She anticipates less of a focus on building out reserving techniques and more of a focus on the “explainability of why the results are the way they are, and how do you actually act on them?”

“I think it’s super exciting, because that’s why people become actuaries, right? You’re not really focused on the data cleansing aspect or the operational aspect. You want to focus on the insights and the next steps.”

She sees AI as “a multiplier for brokers and actuaries,” saying that technological innovation is “just a very exciting opportunity to get more integrated with AI, use it more on a day-to-day basis, and just because it changes your role it doesn’t mean it’s something to shy away from.”

Sure, the actuarial profession needs to adapt and change with the times, but it’s not exactly on weak footing. The Bureau of Labor Statistics sees the number of actuarial jobs growing by 22% between 2024 and 2034, “much faster than the average for all occupations,” according to the agency.¹⁰

In fact, there’s reason to believe that the next five to 10 years wind up being some of the most exciting in the profession’s history, as actuaries are freed up to do more risk architecture and big-picture strategy.

The February sell-off priced in the death of a business process, not the death of the insurance broker business. And while declines in many leading broker stocks were steep, markets didn’t even glance under the hood at how the headline event — an insurer offering quotes directly in ChatGPT — quickly ran into usability speedbumps.

And while ChatGPT may be learning how to talk shop, it’s the human architect who still has to navigate the world that the policy actually protects.