Most of the reinsurance conversation is about price and capacity. Less of it is about the quality of the exposure data carriers hand their reinsurers. ZestyAI calls the cost of that gap an "uncertainty premium": carriers paying for what reinsurers can't verify, not just for the risk itself.
We put our questions to Kumar Dhuvur, Co-Founder and Chief Product Officer of ZestyAI, about that premium, the roof-age problem, what is bringing capacity back to California, and what regulators ask about AI risk scores. His answers are below, lightly edited.
You're calling it an "uncertainty premium." In plain terms, how much more are insurers paying reinsurers because their property data is incomplete or stale, and how did you measure it?
The uncertainty premium is the cost of pricing risk that can't be fully verified. When a carrier's exposure data is incomplete, outdated, or inconsistently coded, reinsurers may have to rely on conservative assumptions rather than the actual characteristics of the properties being insured. That can inflate modeled probable maximum losses (PMLs), influencing reinsurance rates, attachment points, and available capacity.
The financial impact depends on the portfolio, the peril, and how those data gaps affect catastrophe model outputs. Our research examines where that uncertainty enters the reinsurance transaction, from property information collected during underwriting to the exposure files ultimately submitted to reinsurers. In many cases, verified property characteristics never make it into those files, leaving reinsurers to evaluate portfolios using incomplete information or conservative defaults.
In one carrier example, incorporating verified secondary modifiers brought modeled hurricane losses into alignment with actual experience, while revealing previously underestimated wildfire exposure.
The value of better data isn't simply a lower reinsurance premium. It's an independent, third-party view of risk and exposure, rather than relying only on the carrier's own records. That external validation reduces uncertainty and sharpens PMLs, so carriers and reinsurers are more certain of the exposure they're pricing. In some cases, that means identifying previously underestimated vulnerabilities; in others, it means demonstrating that a portfolio is stronger than conventional assumptions suggest. Either way, greater precision enables insurers and reinsurers to price and manage exposure with more confidence.
What are the most common data gaps you see in a carrier's book (roof age, construction, defensible space, something else), and which one moves the reinsurance price the most?
Roof age is one of the most common and consequential gaps we encounter, particularly because insurers have historically relied on estimates from policyholders or agents rather than independently verified information. Our research found that 43% of homeowners understated their roof's age by at least three years, while another 24% overstated it. That creates opportunities for both underpricing and overpricing risk.
But the problem extends beyond roof age. Roof condition, construction characteristics, vegetation proximity, and defensible space are all important attributes that may be missing, inconsistent, or outdated. Much of this information is collected at binding but never refreshed, even as roofs deteriorate, vegetation grows, and homeowners add swimming pools, trampolines, or solar panels.
As for which attribute moves reinsurance pricing the most, there's no universal answer. It depends on the peril, the portfolio, and the catastrophe model being used. Roof characteristics can be particularly important for wind and hail, while vegetation and defensible space can significantly influence wildfire vulnerability.
Ultimately, the greatest value comes from understanding how these characteristics interact and ensuring that verified, current information reaches the catastrophe model rather than relying on broad assumptions.
When a carrier closes those gaps, does the reinsurer actually give them credit at renewal, or does better data mostly help the carrier's own underwriting?
Yes, reinsurers give credit when a carrier's enriched exposure data demonstrates lower modeled risk. When verified property characteristics replace conservative defaults in the exposure file, modeled losses reflect the actual portfolio, and that can improve pricing, structure, and capacity at renewal.
But it works both ways. Verified data can also reveal risk that was previously underestimated. That's what makes it credible: it's an independent, external view of exposure, not a case built only from the carrier's own records. Reinsurers can rely on it because it isn't designed to produce a particular answer.
The carrier benefits well beyond reinsurance. The same data strengthens underwriting and pricing on both new business and renewals, so carriers are selecting and renewing the best risks and capturing the appropriate premium for each one. That includes catching changes since a policy was written, like a newly added swimming pool or solar panels. Getting the full benefit means implementing the data state by state, with the required regulatory approvals in each state.
Better data helps the carrier write a stronger book and gives the reinsurer a clearer view of what it's covering. Both sides benefit.
Capacity is starting to come back into California. What's different about the carriers writing wildfire-exposed business now versus the ones that pulled out?
The biggest difference is the growing recognition that insurers need a more precise understanding of wildfire risk to write that business sustainably.
Historically, insurers relied heavily on broad geographic classifications and backward-looking loss experience. Those approaches can obscure meaningful differences between properties, particularly in wildfire-exposed communities where construction characteristics, vegetation, defensible space, and mitigation measures significantly influence vulnerability.
Today, insurers have access to much more granular property intelligence and sophisticated wildfire models. They can differentiate risks at the individual-property level rather than treating entire neighborhoods as though they have uniform exposure.
We're already seeing that shift. Nine of the 12 carriers that switched models in California since 2025 selected ZestyAI.
California's Sustainable Insurance Strategy is also enabling greater use of forward-looking catastrophe models. Together, these developments create an opportunity for insurers to return with a more disciplined, defensible approach to underwriting wildfire-exposed properties, rather than reverting to the practices that contributed to the market's challenges.
What has to be true on regulation, rate adequacy, mitigation, or modeling for that returning capacity to stick around after the next bad fire season?
For returning capacity to be sustainable, four things need to work together: regulation that supports forward-looking risk assessment, rates that adequately reflect exposure and reinsurance costs, meaningful recognition of mitigation, and models that are continuously validated against actual losses.
California's recent regulatory reforms are an important step, but access to better models is only part of the equation. Insurers must also be able to distinguish between properties that have invested in risk reduction and those that haven't. Otherwise, broad assumptions can continue to penalize lower-risk properties while obscuring greater vulnerabilities elsewhere.
Rate adequacy is equally important. If insurers cannot price the risk they're assuming, particularly as catastrophe exposure and reinsurance costs evolve, the pressure to restrict capacity will return.
The real test isn't whether capacity returns during favorable conditions. It's whether insurers can maintain that capacity after the next major wildfire. That requires a regulatory framework and risk models that support accurate, transparent, and sustainable decisions throughout the underwriting cycle.
Property-level AI models get pushback from regulators and consumer advocates. What's the hardest question a regulator has asked you, and how did you answer it?
A common question we encounter in regulatory reviews is: If an AI-driven risk score affects what a homeowner pays or whether they can obtain coverage, how can that person understand the decision and challenge it?
It's a fair question, and one that gets to the heart of responsible AI adoption in insurance.
Our approach begins with transparency at the individual-property level. Insurers must be able to explain which characteristics influence a property's assessment and how those factors relate to actual loss experience.
Just as important, the score has to be mitigation-aware. If a homeowner has already created defensible space, cleared vegetation, or replaced an aging roof, that should be reflected in their risk assessment. Z-FIRE, for example, recognizes those actions at the individual-property level, so a score isn't a fixed label on a neighborhood. It reflects what's true about that specific home today.
Beyond that, insurers can show homeowners which additional steps would have the greatest impact on their risk. That's what regulators want to see: a clear path to a better outcome, not just an explanation of the current one.
Regulators increasingly expect insurers to provide clear policyholder notices, explain the factors influencing a risk score, and establish processes for consumers to dispute information they believe is inaccurate.
Different states approach these questions differently, which is why we work closely with regulators and carriers to ensure our models and their implementation meet each jurisdiction's requirements.
For an underwriter or agent who isn't a data person, what's one thing they should be asking about the property data behind a quote?
I'd ask one simple question: Where did this property's key risk information come from, and when was it last verified?
Consider roof age. That information might come from a homeowner's estimate, an old application, a building permit, or an analysis of historical aerial imagery. Those sources don't offer the same level of confidence, yet roof age can materially influence pricing, eligibility, and coverage terms.
The same applies to property characteristics that change over time. A homeowner may have replaced a roof, installed solar panels, or made renovations since the information was first collected. Without verification, an insurer may be evaluating a property based on characteristics that are no longer accurate.
Underwriters and agents should know whether the data informing their decisions is accurate, current, and supported by credible sources. Ultimately, even the most sophisticated risk model is only as reliable as the information behind it.
What does your data show about property risk that most of the industry still gets wrong?
One of the biggest misconceptions is that roof age alone is enough to understand roof risk. Age correlates with loss potential, but two roofs of the same age can perform very differently depending on their condition, complexity, materials, and exposure to severe weather.
Our research illustrates just how significant those differences can be. Roofs with 16 or more penetrations experience 2.8 times higher losses than roofs with a single penetration. Similarly, flat roofs or those with 12 or more facets are 55% more likely to generate a claim than simpler two-facet roofs.
There's a maturity curve in how insurers can evaluate that risk. The starting point is accurate roof age, followed by current condition and accumulated damage. Adding roof complexity introduces factors like geometry, slope, materials, and penetrations. The most sophisticated approach examines how those characteristics interact with the frequency and severity of specific hazards.
Advanced machine learning techniques, including gradient-boosted models, can evaluate those interactions together, uncovering relationships that traditional approaches may miss.
The lesson is that risk is multidimensional. Insurers can advance along that maturity curve to make increasingly precise underwriting, pricing, and portfolio decisions.
Kumar Dhuvur is Co-Founder and Chief Product Officer of ZestyAI.