Insuring a more resilient future: How data-driven underwriting is turning risk into market opportunity

When major insurers stop writing property insurance policies in states such as California and Florida, they walk away from multibillion-dollar markets. They base their decisions on seemingly sound actuarial evidence that climate and other risk factors have made entire ZIP codes uninsurable. 

But what if there is a better approach to risk assessment? What if the mass exodus from disaster-prone areas isn't about too much risk, but about too little information?

Applying Google Cloud's geospatial AI to these zones yields a paradoxical result: the pool of insurable properties doesn't shrink — it expands. It turns out that higher elevations or treeless perimeters can give properties better risk profiles than their inland or densely forested counterparts.

In their rush to exit these markets, insurers have been avoiding good risks along with bad ones.

The real cost of data imprecision

Market exits represent more than lost premiums. The ripple effects can destabilize economies because homebuyers can't obtain mortgages without insurance, small businesses can't secure loans to expand, property values stagnate or decline, and local tax bases erode. A vicious cycle takes root.

Another exacerbating factor is that some regulators are forcing insurers to account for climate-related risks in their underwriting and investment portfolios. Retreating from vulnerable markets may satisfy regulators in the short term, but these same regulators will inevitably ask these insurers to help absorb the longer-term economic impact that follows.

The growing "protection gap" — the difference between losses and coverage — threatens to devastate home owners and local businesses who find themselves either completely exposed or paying unsustainable premiums to insurers of last resort. The negative impact cascades through mortgage markets, municipal bonds, and the broader financial system.

AI enables precision analysis

Many insurers are still navigating 21st-century climate complexity using 20th-century risk assessment tools. But now, AI advances are making it possible to analyze geospatial and weather-related data with far more precision, and at far greater scale, than dated maps that eliminate entire regions based on past catastrophic events.

The truth is that property risk varies within a single neighborhood, even within high-risk zones. 

Consider two houses on the same waterfront street. The first house, an original 1960s construction, sits at ground level on a flat lot. Just three doors down, a recent remodel, elevated four feet above grade, features hurricane-impact windows, a reinforced roof system, and natural drainage that channels water away from the foundation.

Traditional regional models classify both properties as uninsurable. But precision analysis reveals that the second house presents far less risk than many properties in "safer" zones. By feeding more granular insurance application data into machine learning models, insurance companies could identify many profitable opportunities in regions they’ve previously abandoned.

Instead of asking, "What zone is this in?" insurers can use AI models to analyze dozens of resilience factors for each individual property: exact elevation, distance to water, roof age and material, presence of storm shutters, defensible space around structures, local drainage patterns, and building code compliance, among others.

Early adopters are already winning

Leading insurers are already using AI-driven analytics to capture market share in territories their competitors have abandoned. They're not taking on more risk, they’re just seeing risk more clearly, and they’re using these insights to identify profitable opportunities and set appropriate prices.

For example, CNA is creating a unified view of environmental risk by combining satellite imagery and vector data (for example, seismic fault lines and property boundaries) with its proprietary information in BigQuery. By performing geospatial analysis directly on this data and using proprietary visualization, CNA can identify subtle but critical factors—like minor elevation changes, flood-prone landscape, and proximity to fault lines and firebreak locations—that traditional assessments miss.

Blyncsy, a Bentley company, is moving beyond static flood maps to provide dynamic, street-level risk assessment. Using AI to analyze near real-time imagery, Blyncsy detects specific trouble spots: stretches of road with persistent standing water indicating poor drainage, debris accumulating at culverts that could cause localized flooding, or deteriorating infrastructure that increases vulnerability. This intelligence transforms a binary "flood zone/not flood zone" decision into a nuanced understanding of precisely which properties face elevated risk and why.

Google Cloud infrastructure powers precision at scale

Achieving property-level precision at portfolio scale requires infrastructure, like Google Cloud, that’s purpose-built for massive data processing and AI deployment:

Google Earth Engine provides a planetary-scale platform for geospatial analysis. Insurers can analyze decades of satellite imagery to understand how properties and their surrounding environments have changed over time. They get a picture not only of current conditions, but also of trends and patterns that signal longer-term resilience or vulnerability.

BigQuery serves as the analytical engine, processing petabytes of climate data, weather patterns, claims histories, and third-party risk factors in real-time. Insurers can run complex queries across massive datasets to model property-specific scenarios at scale. The ability to analyze data in place — without the time and expense of moving data around — accelerates time to insight from weeks to minutes.

Vertex AI transforms data into actionable intelligence. By building and deploying machine learning models that consider hundreds of variables, insurers can estimate loss probability with unprecedented accuracy. These are explainable models that help underwriters understand exactly why a property receives a particular risk score.

WeatherNext, an Earth AI model, provides high-fidelity weather insights at hyper-local resolution, so insurers can understand historical patterns and model future climate scenarios, including potential losses from an upcoming weather event.

Critically, Google Cloud’s AI infrastructure is “Sustainable by Design.” It scales on demand to handle peak analytical loads when necessary while reducing energy consumption and emissions by shifting computing tasks across locations and times of day and using purpose-built AI processors and specialized models.

Turning retreat into opportunity

Just as AI is helping medicine transform from broad disease categorizations to individualized treatment plans, AI-driven insights can help insurance evolve from geographic generalizations to more granular assessment and pricing. By analyzing data at scale, companies can avoid genuinely uninsurable properties and identify properties that older methods inaccurately flag as untouchable.

Insurers who adopt AI-driven approaches will discover that the path to growth doesn't shy away from climate-exposed markets but runs straight through them, guided by data and powered by AI models like Earth Engine and WeatherNext. 

A virtuous cycle replaces the vicious one, with additional data from insurance applications, claims, and continuously updated weather and geospatial information improving outcomes over time. With a better understanding of individual properties, as well as more accurate forecasting through predictive AI, insurers can help their customers get ahead of potentially catastrophic events, thus reducing losses.

The benefits extend far beyond profitability. When risk is understood, companies can insure more properties with higher confidence, and communities can build a more climate-resilient future. When properties remain insurable, communities can invest for the long term. When everyone has a financial safety net, economies grow.

Learn more here.