Simon Glairy stands at the intersection of traditional risk management and the cutting edge of Insurtech, where he has spent years dissecting how technology integrates with high-stakes decision-making. As insurance leaders grapple with the hype and reality of artificial intelligence, Glairy offers a sobering yet optimistic view on why many initiatives fail to cross the finish line of profitability. His perspective is rooted in the belief that AI is not a tool to be bought but a framework to be built around the hard-won wisdom of the industry’s veterans.
Our discussion highlights the critical need for insurers to pivot from viewing AI as a “super-chatbot” to treating it as a dynamic repository for institutional expertise. We explore the specific vertical applications in claims and underwriting where the technology is already paying dividends, the importance of deterministic guardrails in system architecture, and the fundamental danger of letting go of human talent just as it becomes most necessary for refining these systems.
Many firms treat AI as a standard software rollout rather than a fundamental rebuild of institutional expertise; how does this mindset specifically lead to failed pilots?
When executives approach agentic AI as just another piece of software to be installed, they miss the reality that this technology must capture and scale human wisdom to be effective. The most persistent misconception I see is mistaking these sophisticated agents for nothing more than souped-up chatbots designed to handle basic queries. In reality, the value isn’t found in the technology itself, but in how it reflects the unique DNA of the firm’s decision-making processes. If you just roll it out without restructuring how expertise is codified, you end up with a tool that works in a vacuum but fails to compound value over time. You cannot simply plug in a model and expect it to understand the nuance of a specific market without first rebuilding the way that expertise is captured and stored.
Why do you argue that the design choices and orchestration around a model are actually more important for competitive advantage than the raw intelligence of the AI itself?
In my experience, orchestration and contextualization beat raw model intelligence every single time because the architecture determines how a tool actually performs in the real world. You could have two different insurers using the identical foundational model, yet they will see radically different results based on how they structure the agent’s access to tools and the order of operations. It is about the deterministic guardrails you put in place and the way you embed your specific institutional expertise into the system’s workflow. This is where the true competitive advantage lives; it’s the difference between a general-purpose engine and a finely tuned racing machine built for a specific track. Without that careful architecture, a powerful model is just a loud engine that doesn’t know which direction to drive.
As the industry moves past the initial hype, why is the return on investment for general AI becoming harder to prove, and where should leaders look instead for financial gains?
We are seeing that while general-purpose AI tools can certainly boost individual productivity, those small efficiency gains are notoriously difficult to track in a clean-cut way on a balance sheet. The real financial impact—the kind that moves the needle for a global firm—comes from vertical, insurance-specific AI systems that are deeply embedded into core workflows. These systems allow for continuous iteration and scaling, which creates a compounding effect that general tools simply cannot match. Many insurers have already proven the technology works in isolated, “clean” tests, but converting that success into a system that generates long-term returns is a much harder, more specific challenge. To see a real return, you have to move away from horizontal experiments and commit to deep, vertical integration in areas like pricing and risk clustering.
Could you walk us through a specific example where a focused, agentic system delivered quantifiable results in a relatively short timeframe?
One of the most impressive cases I’ve seen involved an insurer working with WTW to transform their claims process, and the numbers were staggering. Within just ten months, they were able to produce over $13 million in annual savings on claims costs while maintaining a 95% accuracy rate for First Notice of Loss decision-making. This wasn’t achieved by replacing humans with a “black box,” but by using agentic systems to handle the intake, assessment, and suggestion layers of the workflow. The system supported the ingestion of submissions and enriched them with both internal and external data, benchmarking risks against portfolio peers in real-time. This level of precision allows for the flagging of coverage misalignments almost instantly, which is where those millions of dollars in leakage are actually saved.
What is the primary human-centric bottleneck that prevents these systems from reaching their full potential?
The toughest obstacle isn’t the code; it’s the difficulty of converting “invisible” human skill into a structured format that a digital system can actually utilize. An experienced underwriter or claims handler carries decades of pattern recognition, market intuition, and contextual judgment that they often struggle to even articulate. This expertise manifests differently across every non-standard case they encounter, making it incredibly hard to write down as a set of rules. If the system cannot tap into that silent intuition, it remains a blunt instrument rather than a sharp tool for risk assessment. We have to find better ways to bridge the gap between what a veteran “just knows” and what an AI agent needs to perform.
There is a lot of anxiety regarding AI and job security; why do you believe that hollowing out the expert base to save costs is actually a strategic mistake?
It is a dangerous gamble because agents are only ever going to be as good as the expertise of the people who are teaching and working alongside them. In a highly regulated industry like insurance, decisions require human accountability, and if you remove the experts, you essentially hollow out the “brain” of your AI. You lose the very source of the pattern recognition and market nuance that makes the technology effective in the first place. I am very skeptical of using AI as a primary reason for workforce reductions because a culture that only consumes technology will always be behind. The successful firms will be those that shift toward shaping technology, which requires keeping your most experienced people at the center of the process.
What is your forecast for the insurance industry as agentic AI moves from the pilot phase to full-scale operational reality?
I believe the industry is heading toward a massive cultural shift where the most successful agentic insurers will stop being mere consumers of technology and start becoming the architects of it. We will see a clear divide between firms that use AI for basic automation and those that use it for genuine augmentation in complex areas like portfolio management and latent risk clustering. In the coming years, actuarial judgment will still drive the process, but it will be supported by agents that handle the heavy lifting of documentation and cross-referencing assumptions in real-time. The firms that win will be those that realize their primary asset isn’t their data or their models, but the way they use technology to amplify the human expertise they already have.
