The Center for Data Innovation recently spoke with Samuel Gold, founder of Risklytics, a San Francisco-based company developing an AI-powered platform that helps insurers and investors assess natural-disaster risk for individual properties. Gold explained how the company’s models analyze property-level data to estimate wildfire risk and support more accurate underwriting decisions.
David Kertai: What does Risklytics offer?
Samuel Gold: Property-level wildfire underwriting remains slow, imprecise, and difficult. Most insurers still rely on hazard maps and catastrophe models designed to estimate risk across large geographic areas or entire insurance portfolios rather than individual properties. As a result, underwriters often treat every home in a high-risk area the same, even though neighboring properties can face very different levels of risk. This causes insurers to overprice or decline many properties that could actually be insured safely.
Risklytics addresses this challenge by evaluating natural-disaster risk one property at a time, beginning with California wildfires. Our platform combines AI models, large-scale simulations, and detailed property data to generate a property-specific risk score and estimated financial loss. Insurers, reinsurers, and managing general agents—specialized firms that underwrite policies on behalf of insurance companies—use these insights to price and select risks more accurately. Investment firms also use the platform to assess disaster exposure across real-estate portfolios.
Kertai: What types of data does your platform analyze to assess catastrophe risk?
Gold: Our platform combines satellite and aerial imagery, environmental conditions, property characteristics, topography, vegetation, weather data, and historical disaster losses. The AI models continuously update and organize these datasets as new information becomes available, ensuring the models reflect current conditions rather than outdated snapshots.
The real advantage comes from integrating these diverse data sources into a single proprietary dataset. By analyzing how they interact, the platform develops a much more complete picture of each property’s exposure than any single dataset could provide.
Kertai: How does your platform simulate the impact of natural disasters on individual properties?
Gold: Rather than assigning every home in a neighborhood the same level of risk, we create a digital model of each property and simulate how it performs under thousands of potential wildfire scenarios. The simulations account for factors such as building characteristics, surrounding vegetation, terrain, weather, and other environmental conditions that influence fire behavior.
The system then estimates both the likelihood of damage and the expected financial loss for that property. This gives insurers and investors a much more detailed understanding of risk than broad regional models and helps them make underwriting and investment decisions with greater confidence.
Kertai: How do your AI models improve on traditional catastrophe-risk models?
Gold: Traditional catastrophe models help insurers understand the overall exposure of an insurance portfolio, but they provide limited insight into whether a specific property represents a good underwriting opportunity. Our platform starts with the individual property and builds upward, allowing insurers to distinguish lower-risk homes from higher-risk ones within the same region.
The platform also identifies the factors driving a property’s risk, such as nearby vegetation or building characteristics, and shows how mitigation measures can reduce that risk. This helps insurers cover properties they might otherwise reject while giving homeowners clearer guidance on improving their property’s insurability.
As regulations evolve, this level of transparency is becoming increasingly valuable. Some states are beginning to require insurers to explain property risk scores and show how mitigation efforts affect them. Our platform already provides that level of detail, helping customers prepare for emerging regulatory requirements.
Kertai: What challenges have you faced in building accurate AI models for natural-disaster risk?
Gold: One of the biggest challenges is that catastrophic events occur relatively infrequently, which means high-quality training data is limited. We must build models that remain accurate and reliable even when historical examples are sparse. That requires extensive validation against real wildfire events and continuous testing as new data becomes available.
Another challenge is earning users’ trust. Underwriters need more than a risk score, they need to understand why the model reached its conclusion. We focus on producing stable, explainable results that clearly identify the factors driving risk, allowing insurers to make confident underwriting decisions. Ultimately, our goal is to give the insurance industry more precise tools to expand coverage while improving resilience in disaster-prone communities.
