The Data-Driven Future of Insurance: AI Agents, Insurtechs & the Battle Over Profits
Two weeks ago, I posted about Lemonade, and it racked up over 250,000 impressions. So, needless to say, I went viral. 😊 Okay, maybe not viral, but it’s clear that people find the topic fascinating. A lot of that interest comes from a collective fear—will AI take our jobs?
This time, I want to shift the focus to what can go right with AI in insurance. Because like it or not, AI is going to be a major part of our daily lives moving forward.
NVIDIA’s AI Vision: Bigger, Faster, Stronger (For AI Agents & GPUs)
NVIDIA CEO Jensen Huang just delivered his keynote at the GTC Conference, laying out his post-DeepSeek projections on AI’s future—and, unsurprisingly, NVIDIA’s role in it. Now, take this with a grain of salt (he does have chips to sell), but Huang predicts that with the rise of agentic AI—autonomous AI agents capable of executing tasks with minimal human intervention—GPU demand will be 100x higher than estimated just a year ago.
This is a bit at odds with DeepSeek’s approach, which aims to reduce the reliance on GPUs, but Wall Street seems to be backing NVIDIA’s vision. Morgan Stanley, Wells Fargo, and JP Morgan all signaled their support for the outlook.
AI Agents: The Next Step in Insurance Automation?
Every major brokerage and carrier is testing AI products, and if you aren’t, you’re already behind. But here’s the reality: no one has truly built a bespoke AI agent for insurance brokers—yet.
Why? Regulation, privacy concerns, and legacy system integrations.
Most brokers work within either AMS or Epic/Applied, and for an AI agent to be useful, it needs seamless integration with these systems. That requires a level of trust, safeguards, and compliance that just isn’t there yet.
Right now, AI tools exist for policy review, comparisons, and certifications, but they’re not truly transformative. Most of what’s available can already be done with ChatGPT 4.0 or similar models. The real game-changer will come when we can offload underwriting, quoting, and risk analysis to an AI agent with a single prompt.
Lemonade’s AI Play: Has It Paid Off?
Ah yes, my favorite punching bag.
Lemonade likes to tout AI-powered underwriting, but in reality, they’ve mostly automated the submission and underwriting process for personal lines. And, as I pointed out before, it hasn’t worked out so well. Their loss ratios remain high, regulatory scrutiny is constant, and they still struggle to prove that AI-driven underwriting is superior to traditional methods.
Some people have argued that Root Insurance has been more successful. But Root isn’t exactly using AI-driven underwriting either—it’s using telematics. By tracking driving behavior via smartphone GPS and accelerometers, Root adjusts rates dynamically based on real-world driving habits.
Tesla: The Poster Child for Adverse Selection?
Tesla’s insurance model is similar to Root’s. Their initial thesis? They own the vehicle, they own the data, so they should be able to price coverage better than anyone else.
The problem? They didn’t account for who their customers actually are.
Excluding the current vandalism claims issues, Tesla owners have proven to be riskier than average drivers—skewing heavily male, getting into larger accidents, and exhibiting higher-risk behaviors. The result? A case study in adverse selection.
It turns out that owning the data doesn’t automatically mean better underwriting. Tesla may very well be the poster child for a failed insurtech experiment.
The Bottom Line: Where Are Insurance Profits Headed?
The battle between human underwriters vs. AI-driven models is far from settled.
Traditional insurers make money through underwriting discipline—carefully selecting risks and pricing them appropriately. AI tries to replicate that but lacks the one thing that’s defined the industry for centuries: human intuition.
For 400 years (+/-), insurance has been a relationship-driven business. The most successful underwriters have always been those who can judge character, assess risk intuitively, and occasionally take a strategic bet on a high-premium, high-risk, high-reward account.
If you look at profitability across books of business, it’s often the unexpected accounts that end up being the most profitable—big premium, few losses, and long-term stability.
So, What Does This All Mean?
Honestly? I don’t know. But what I do know is that you shouldn’t take your foot off the gas.
Keep investing in AI. Keep testing AI. But remember that the human element is what has made insurance one of the most enduring industries in history.
Let’s hear it—where do you see AI and insurtech taking the industry next?