How artificial intelligence is reshaping insurance ecosystems and forcing the industry to evolve

An 88-year-old Miami transplant named Helen represents both the unlikeliest and the clearest example of how artificial intelligence is completely upending the relationship between insurers and the insured. 

Helen starts every morning before dawn walking with ankle weights and neon flashing athletic shoes. After her walks, she pours herself what she calls “North Carolina coffee,” her code for coffee spiked with moonshine. She was born and raised in Asheville, where her entire family still lives.

In September 2024, Hurricane Helene devastated her hometown in North Carolina. Helen showed up at her neighbor’s door holding back tears. Many of her family members had lost properties because they were without flood insurance. Who would have predicted catastrophic flooding in the mountains of western North Carolina?

The neighbor Helen poured her heart out to happened to be Astrid Malval-Beharry, founder of StratMaven, a boutique strategy and M&A advisory firm focused on the intersection of P&C insurance and technology. Helen wanted help reviewing her own policy. Instead of flipping through documents the way most industry professionals might, Malval-Beharry tried something different.

She scanned Helen’s current policy and renewal documents and uploaded them to a generative AI tool. Malval-Beharry wrote a quick prompt assigning the AI chatbot the role of an insurance agent. She requested that the “AI insurance agent” explain in plain English what had changed between policies. She told it to  flag any concerns and suggest questions Helen should ask her actual human insurance agent.

In less than two minutes, they had clear, comprehensive answers. It was exactly the  kind of clarity Helen had been seeking from her agent for three decades.

Helen, with characteristic tartness, called her agent back in Asheville. “Mr. Fairbanks,” she said, “I just learned this new thing called ‘artificial intelligence.’ You may not be aware of it. But I am now. It did for me in two minutes what I’ve been asking you to do for the past 30 years. Either you get on with the program, or I’ll find myself a new agent.”

For Malval-Beharry, the moment crystallized something critical that some carriers and vendors still don’t fully realize. Generative AI isn’t just for automating back-office processes anymore.  It’s a tool that works for everyone, and it’s creating what she calls “AI-empowered policyholders.”

“Once consumers like Helen, or anyone really, experience AI’s speed and clarity, they won’t go back,” she says.

Over the past 25 years, it’s become a cliché to gamely acknowledge that “the consumer is in control.” The thing is, now, that truism is more real than ever. 

Getting insurance professionals and organizations to accept this newly democratized resource is something Malval-Beharry has been working on her entire career. Her unusual introduction to insurance cast her in the role of industry iconoclast from the get-go. 

From Victoria’s Secret to insurance strategy

About 99% of insurance professionals “fall into” the job, Malval-Beharry jokes. As she tells it, Malval-Beharry faced a simple choice: work in insurance or on Victoria’s Secret. To her, it was an easy pick.

It was the early 2000s and Malval-Beharry was working in Boston Consulting Group’s New York office. She could either  join the Victoria’s Secret project that everyone at BCG seemed to be vying for… or take on an insurance engagement that nobody wanted. With an engineering background and self-described data obsession, she chose insurance and never looked back.

“I fell in love with the industry and its mission,” she says. 

Her career path took her through carrier consulting work and a P&L role at LexisNexis overseeing the homeowners insurance vertical, where she learned to solve carrier pain points through technology, data, and analytics. 

Thirteen years ago, she founded StratMaven, where she and her team help carriers, MGAs, technology vendors, service providers, and investors make better strategic decisions and turn them into execution, across growth, digital transformation, and competitive challenges.

Naturally, AI and digital readiness are taking a lot of the team’s focus these days.

The transformation is already happening

From her work across the insurance ecosystem, Malval-Beharry has witnessed boards putting intense pressure on C-suite professionals to develop AI strategies while regulators race to catch up. 

But what’s different now, she says, is that in just the past 12-to-18 months, several carriers have moved insurance-specific large language models into production.

“That kind of speed is absolutely unprecedented in our industry,” she notes. “The typical sales cycle in insurance is 12 to 18 months. To go from pilot into full production in sometimes six months brings a whole new set of challenges to organizations and staffers.”

ChatGPT and other LLMs have democratized all kinds of time-consuming processes and the ability to synthesize enormous amounts of data. It’s true that machine learning programs and traditional AI have already delivered better predictions and tools for producing hazard scores, roof condition assessments, propensity-to-buy metrics. But generative AI and agentic AI are changing the scope of what software can do in three fundamental ways, Malval-Beharry says.

First, these systems understand and generate language at a human level. In an industry that runs on documents, spreadsheets, CSV files, emails, and ACORD forms, this matters significantly. Instead of armies of people reading PDFs and re-keying data, large language models can now ingest and summarize broker submissions, loss runs, and inspection reports. They can draft underwriting rationales. They can even answer phones with an approximation of genuine empathy.

Second, AI can now orchestrate multi-step workflows rather than just scoring single events. 

As Malval-Beharry notes, traditional AI might flag a claim with a probability score for fraud. Agentic AI can take a new notice of loss, pull policy details and prior claims, enrich them with third-party data, classify and triage the claim to the right adjuster, schedule an inspection, and loop in humans at appropriate thresholds.

“We are moving from AI that advises and gives a score, to AI that actually acts, albeit within necessary guardrails,” she explains. “That latter part, setting up guardrails, is increasingly essential. And it’s often overlooked, to organizations’ regret.”

Third, AI is also democratizing the way insurance organizations access advanced analytics capabilities. 

Throughout the modern history of insurance, only the largest tier-one carriers could afford armies of data scientists and massive tech stacks. Now, small MGAs (managing general agents) are leveraging large language models without needing that infrastructure. It hasn’t quite flattened the playing field there, Malval-Beharry says. But it’s starting to level it for those willing to change how they work.

Three speeds of transformation

Malval-Beharry points to “three speeds of AI adoption” happening simultaneously across the industry.

The fast track involves low-hanging fruit like document ingestion and straightforward communication tools. “There’s a new deployment every week,” she says. “You’re moving from pilots to production in weeks, not months or years.”

Medium-speed adoption involves higher-value tasks where organizations have established trust in models and processes. AI is being rolled out across workflows involving pricing, reserving, and claims decisions — triaging first notice of loss, applying underwriting rules to submissions, extracting data and preparing files for human review.

The slower transformation involves fully autonomous — with human oversight — end-to-end decision-making. This means AI agents that don’t just service but reason, orchestrating entire workflows. She expects this to deploy over the next three-to-five years.

“The best carriers and service providers will be the ones that quietly redeploy human time towards judgment, relationships, and complex problem-solving,” she says, “while integrating AI agents into processes to handle orchestration and drudge work.”

Existential questions — and threats — for service providers

For service providers, whether business process outsourcers or third-party administrators handling claims, Malval-Beharry offers a clear warning. AI isn’t going to replace services, but it will redefine what services mean.

A recent conversation with a carrier’s head of AI stuck with her. 

The executive observed that no one ever asks about having “machines in the loop” to protect against human error, something that happens constantly. That reframing, she argues, is critical. It all comes down to how service providers can use AI to augment what humans do best.

“If you are a service provider today and your value proposition is essentially ‘We have smart people and lower-cost labor,’ then agentic AI is an existential threat,” she warns.

Service firms must reposition from labor providers to capability partners. She identifies four strategic moves for survival and success.

The first is shifting from selling hours or full-time equivalents (FTEs), which allows employers to measure staffing numbers,  to selling outcomes. Instead of promising 50 FTEs, commit to specific cycle times, premium leakage reduction, or claims efficiency metrics. AI agents become part of delivering on that promise rather than a threat to it.

Second, service providers should build and productize proprietary playbooks. A TPA can develop proprietary claims workflows, configuring AI agents and humans in ways that differentiate them from generic providers.

Third, become an orchestrator of human and digital work. Service firms are uniquely positioned to design AI-first processes with clear swim lanes for humans and agents. But this requires rethinking target operating models. It’s not just taking client processes and running them cheaper. Instead, Malval-Beharry says, it calls for truly re-architecting for an agent-enabled world.

Fourth, invest in talent that can supervise AI rather than compete with it. This might mean upskilling existing staff or creating new roles like “AI workflow supervisor.” Some firms are taking minority investments in agentic AI companies, turning technology providers into genuine strategic partners.

Ecosystems that actually work

No single vendor, carrier, MGA, or service provider can match the current pace of change, Malval-Beharry argues. Modern transformation efforts require core system upgrades paired with third-party data providers, AI agents, TPAs, and workflow platforms. No single player can credibly build all those capabilities and keep them best-in-class.

Partner ecosystems aren’t new; core system vendors like Guidewire have long maintained partner networks enabling API-based integrations, Malval-Beharry notes. But because of how agentic workflows operate, these partnerships must now go beyond press releases.

The ecosystem partnerships that work best share three traits, she says. First, there’s clear ownership of the customer relationship. Everyone knows who owns the end customer and who provides infrastructure. When something breaks, accountability is defined.

Second, there’s transparent data and value sharing, which include explicit agreements about what data moves where, how it can be used, and how economic value is distributed.

Third, too many organizations ignore the work of “joint execution,” she says. It’s one of the primary issues that have held back insurtech investments and programs.

“They treat the ecosystem as a product, not a procurement exercise where you check a box,” she explains. Successful partnerships, in Malval-Beharry’s view, assemble capabilities quickly enough to keep pace with AI evolution and new data sources.

Building for lunchtime

In the first wave of insurtech, most capital came from generalist VCs and growth equity funds. The narrative was premised on disruption. The aim was to replicate in insurance what FinTech achieved in banking. Carriers were often pitched as the problem to be solved, Malval-Beharry says.

Today, approximately 40% of Insurtech funding comes from insurers and reinsurers themselves, up from 5-10% in that first wave, she estimates. “I wouldn’t be surprised to see that figure reach 50% by the end of 2026,” Malval-Beharry adds.

That expansion reflects a number of changes. Carriers and reinsurers are now major limited partners and direct investors, meaning they’re not just buyers of technology but architects of it. Insurtechs with insurer investors tend to perform better because those investors shape product roadmaps to align with real carrier pain points rather than generic tech narratives.

“The relationship has shifted from adversarial to symbiotic,” Malval-Beharry says. “Insurtechs learned they can’t disrupt insurance from outside without deep domain and regulatory expertise. Many investors now won’t back founders without insurance veterans on the team or clear timelines to bring them in.”

Capital now follows demonstrated ROI rather than giant total addressable market slides. Insurtechs attracting insurer capital tend to solve specific high-value problems in underwriting, claims, distribution, or risk management. They have production deployments, not just pilots, and they integrate into core systems—making them sticky.

“Build for lunchtime,” Malval-Beharry advises insurtech founders. “If your solution were to disappear, would the carrier notice by lunchtime or next month? You want to be so essential that switching costs are elevated.”

Insurtech is no longer a parallel ecosystem. It has become part of the industry’s “self-disruption mechanism,” Malval-Beharry says. “When 40 cents of every insurtech dollar comes from insurers and reinsurers, you’re no longer looking at competitors, you’re looking at builders.”

The cynicism pitfall

“Hype-cycle” warnings have never been so intense as in the case of AI’s rise. The talk of an “AI bubble” is everywhere. Fears over practical rewards and promised efficiencies hover over the C-suite on down.

But Malval-Beharry has little patience for AI cynics. “Dismissing AI because of the hype is like dismissing the internet in 2000 because of the dot-com bubble,” she says. “Yes, there may be a data center buildup bubble. Yes, models can hallucinate. But the underlying transformation is real, just as it was for the internet.”

Hype, to her, means claims that AI will replace all underwriters and adjusters in two years, or promises to transform entire operations in six weeks, or slapping “AI-native” on platforms without substance, or making promises without unit economics to back them up.

She points to a Conning study published in June 2025: large language model adoption among insurers jumped from 3% to 32% in a single year. “That’s actually crazy,” she acknowledges. “But it’s also real.”

Her advice is simply to do your due diligence. Be skeptical of vendors. And demand proof. Ensure models are transparent, explainable, and auditable. “If results aren’t returned in minutes, that’s not generative AI, that’s someone typing on a keyboard,” she says.

The carriers winning right now are cutting through the noise, identifying real use cases backed by data, and executing with discipline. “If you don’t keep up, you’ll be left behind,” Malval-Beharry says. “Bubble or not, of that I have zero doubt.”