Simon Glairy is a recognized leader in the Insurtech space, specializing in the intersection of risk management and AI-driven behavioral assessment. With years of experience helping carriers navigate the shift from traditional underwriting to digital-first distribution, he has become a leading voice on how to balance rapid customer acquisition with long-term profitability. As digital insurance platforms become more sophisticated, Glairy focuses on “Behavioral Intelligence”—a method of analyzing how applicants interact with digital forms to detect intent and risk before a policy is ever bound.
In this conversation, we explore the evolving landscape of premium leakage, the rise of “self-underwriting” behaviors among digital shoppers, and how carriers can use real-time micro-behaviors to neutralize fraud without damaging the user experience.
Streamlining quote-and-bind experiences with prefill and fewer questions often improves conversion but may increase customer switching. How does this lack of friction contribute to premium leakage, and what specific impacts have you seen on loss ratios when applicants manipulate data like mileage or garaging addresses?
The drive for a friction-less experience is a double-edged sword because when you make it easier for a customer to buy, you also make it easier for them to misrepresent the truth. We are seeing a significant rise in premium leakage, which was estimated at roughly $29 billion annually in personal auto back in 2016, and given inflation and premium hikes, that number is likely much higher today. When applicants realize they can lower their premiums by “gaming” the system—such as failing to disclose a youthful driver or moving a garaging address to a cheaper ZIP code—it results in loss ratios that are 50% to 500% higher than the baseline. It is a sobering reality that 40% of millennial applicants admit to using these deceptive tactics, creating a massive gap between the risk the carrier thinks they are taking on and the actual exposure.
Many applicants effectively underwrite themselves by editing fields or removing youthful drivers after seeing premium changes. Why are traditional underwriting models unable to capture these “in-session” behaviors, and how do real-time signals like typing speed and hesitation reveal the difference between honest mistakes and deliberate rate evasion?
Traditional underwriting is inherently retrospective; it looks at the final data submitted and checks it against static sources like credit scores or MVRs, but it misses the entire journey of how that data was created. We call this the “self-underwriting” problem, where an applicant toggles different mileage brackets or deletes a prior violation only after seeing the premium jump. By capturing thousands of digital micro-behaviors, such as hesitation, excessive field edits, or copy-paste activity, we can see the “why” behind the data. For instance, a long hesitation on a specific question followed by multiple changes to a numeric value often signals deliberate rate evasion rather than a simple typo, allowing carriers to see the intent beneath the surface.
Organized fraud rings and automated tools now exploit digital application flows alongside standard consumers. What specific digital micro-behaviors, such as navigation patterns or device anomalies, signal an automated attack, and how can carriers neutralize these threats without introducing broad friction that harms legitimate high-intent shoppers?
Fraudsters have modernized their tactics just as fast as carriers have modernized their distribution, often using headless browsers or automated AI tools to “scrape” or exploit application flows. We look for signals like non-human navigation patterns, incredible typing speeds, or device anomalies that suggest a bot rather than a person is filling out the form. To neutralize these without hurting the average shopper, carriers can use “surgical friction”—where only the sessions displaying these specific high-risk behaviors are flagged for manual review or additional verification. This allows the 95% of legitimate, high-intent shoppers to breeze through the process while effectively putting a “digital speed bump” in front of the bad actors.
High purchase intent can sometimes mask a high-risk profile, such as someone seeking immediate coverage after an accident. How do behavioral signals help distinguish between these two scenarios, and what are the practical steps for monetizing low-intent traffic while protecting the integrity of the core book?
This is a critical distinction because a “highly motivated” buyer isn’t always a “highly desirable” customer. Someone who just bought a new car and someone who just had an uninsured accident both show high conversion intent, but their risk profiles are polar opposites. Behavioral intelligence looks for patterns of urgency or specific data manipulation that might suggest an applicant is seeking immediate coverage for a pre-existing loss. For those who show low intent or high risk, carriers can strategically route them to “click listings” or monetization partnerships. This allows the carrier to recoup acquisition costs through referral fees without actually taking the risky policy onto their own books.
Carriers often swing between prioritizing growth and margin protection by adding or removing manual reviews. How does embedding behavioral intelligence allow for a more surgical approach to friction, and what are the long-term strategic advantages of dynamically routing users based on their real-time risk profile?
Historically, when loss ratios climb, carriers “slam on the brakes” by adding more questions or disabling digital binding for everyone, which kills growth and frustrates good customers. Behavioral intelligence ends this “pendulum swing” by allowing for dynamic calibration where the application experience changes based on the user’s real-time risk profile. Clean, honest applicants get the “fast track” experience with prefilled data and instant binding, while high-risk sessions are routed to agents or require extra documentation. Long-term, this creates a massive competitive advantage because the carrier can keep their “growth engine” running even in tough markets by precisely filtering out the “bad” growth.
Beyond the initial application, behavioral data is being used in endorsements and claims. What specific patterns indicate risk during the First Notice of Loss (FNOL) phase, and how does integrating these insights across the entire policy lifecycle help reduce the overall cost of acquisition and servicing?
The power of behavioral data extends far beyond the quote; it is now being applied to endorsements, servicing, and especially the First Notice of Loss (FNOL) phase. During a claim, we look for patterns of hesitation or “story-building,” where a claimant might change the details of an accident multiple times before submitting the digital form. By identifying these signals early, carriers can fast-track simple, honest claims to improve customer satisfaction while assigning suspicious claims to specialized investigators. Integrating these insights across the lifecycle reduces the “total cost of risk” by ensuring that the carrier isn’t just acquiring customers efficiently, but is also managing the ongoing integrity of the policy through to the claim.
What is your forecast for digital insurance?
I believe the future of digital insurance will move away from the current “one-size-fits-all” model and toward a fully dynamic, “intelligent” application environment. We will see a shift where the application itself becomes a living tool that adjusts its difficulty and verification requirements in real time based on the user’s behavior. Carriers that continue to rely solely on static, third-party data will find themselves “adverse-selected” by savvy consumers and automated fraud tools that know how to hide between the data points. Ultimately, the winners in this space won’t be the companies that build the highest walls, but those who use the most sophisticated intelligence to let the right people in as quickly as possible.
