Risk Assessment & Property Intelligence

From snapshots to signals: the rise of real-time risk intelligence 

BCG’s AI Radar shows that industry AI spending as a share of revenue will triple in 2026, yet just 38 percent of P&C insurers are generating value at scale from AI in core workflows. This gap is deeper than technicalities – it’s structural and firmly rooted in an industry that has spent decades assessing property risk on an annual cycle, then filing the results away until next renewal. 

The problem is that risk doesn’t pause between renewals. Vegetation encroaches, drainage infrastructure ages, land use shifts around a property. A risk that scored acceptably 12 months ago may look materially different today. Under a static assessment model, nobody knows until a claim arrives because what the industry has historically worked with is not the same as what is now available. What’s more, advances in commercial earth observation data have unlocked analytics that are moving underwriters beyond static, point-in-time assessments toward a dynamic picture of how conditions on the ground are evolving, from seasonal vegetation shifts to post-event change detection.  

From a market standpoint, these advancements are accelerating - however adoption still remains uneven. Only around 30 percent of insurers are currently using geospatial and location intelligence in underwriting - meaning a large share of underwriting decisions are still being made without a clear understanding of the risk involved.  

But having that data is only half the challenge. Successful adoption depends on three things working together: the data must arrive fast enough to support a decision at the point of bind; it must reach underwriters inside their existing decision-making workflow; and it must be explainable – when a carrier declines a risk or applies a surcharge based on a geospatial signal, it needs to answer what was observed, where, when, and why it is material.  
 

“The central challenge is that the ‘average’ view of risk no longer holds. Catastrophe losses – wildfire, flood, severe convective storm – are rising in both frequency and severity, and exposure keeps concentrating in higher-hazard locations, pushing prices up and, in some markets, pulling capacity out altogether”
Richard HartleyCytora 

The climate data problem

The numbers from 2025 reframed what the US insurance industry thought it understood about catastrophes. Total global insured losses for the year reached $129 billion, with the US accounting for $100 billion, 78 percent of the global total. The January Los Angeles wildfires alone generated an estimated $41 billion in insured losses, while severe convective storms were responsible for at least 47 percent of total insured losses worldwide, equating to $60 billion globally and $51 billion in the US.  

And these events are reshaping everything the industry knew about risk.  

“The central challenge is that the ‘average’ view of risk no longer holds,” Richard Hartley, co-founder at Cytora, told Insurance Business. “Catastrophe losses – wildfire, flood, severe convective storm – are rising in both frequency and severity, and exposure keeps concentrating in higher-hazard locations, pushing prices up and, in some markets, pulling capacity out altogether. AI and deep learning are reshaping catastrophe modeling in response, moving beyond coarse, static hazard maps toward higher-resolution, property-specific views of flood and other perils, and wider megatrends in weather, technology, and data are forcing a rethink of how firms quantify physical risk.” 

The catastrophe modeling infrastructure that underpins US property underwriting was largely constructed around peak perils that are high-severity, geographically concentrated, and supported by decades of historical loss data. Moody’s analysis confirmed that severe convective storms produced more than $45 billion in insured losses in the US for the third consecutive year, pointing to urban sprawl, rising repair and reconstruction costs, and social inflation as key amplifiers.  

The conclusion? Severe convective storms must now be treated as a primary, not a secondary, peril in insurers’ portfolios.  

And so, the response from the modeling industry is also accelerating. Gallagher Re’s 2026 First View Report identified the expanding use of tech as a key development in weather and climate forecasting, with AI and technology now incorporated into official agency tools used to predict tropical cyclone formation, track, and intensity – though further refinement is still required to improve accuracy for precipitation and other compound drivers of storm risk.  

“Technology helps in three ways,” says Hartley. “Granularity – enriching every submission with property-level intelligence (construction, occupancy, geospatial, and climate data) so insurability rests on the specific risk, not a postcode average; speed – digitizing and triaging CAT-exposed submissions automatically so scarce expertise is reserved for the genuinely complex risks; and consistency – applying the same appetite and accumulation rules to every risk, every time, which matters enormously when managing aggregation in peak zones.” 

[INSERT INFOGRAPHIC 1]

AI and satellite imagery: property intelligence gets granular 

For most of the industry’s history, property risk assessment meant what an underwriter could see on a form: address, construction type, year built, occupancy. What they couldn’t see was the actual condition of the property. Whether the roof had deteriorated since last renewal, whether any wildlife had moved in, whether an unpermitted outbuilding had appeared on the lot – that information existed on the ground. Getting to it reliably was the problem – and now AI and satellite imagery are solving it. 

The most visible application is aerial imagery analysis – vendors fly fixed-wing aircraft and drones, pull satellite photos, and feed the images to models that score roof condition, vegetation proximity, debris, pools, trampolines, and unpermitted structures. What was once reserved for high-value commercial risks – a physical inspection to verify what the submission actually said – is now executable on the entire book, automatically, at renewal. Computer vision tools translate raw imagery into structured data, such as roof condition scores, defensible space metrics, and hazard indicators, that underwriters and claims teams can use at scale, making aerial imagery not just a visual aid but a foundational data source in modern property insurance operations.  

Regulators, however, are paying close attention here too. State insurance departments emphasize that aerial imagery should typically be one input among several, not the sole basis for cancelations, nonrenewal, declinations, or significant rating changes. Colorado’s Division of Insurance issued Bulletin B-5.57 in March 2026, warning that images requiring significant enlargement or inference may need verification by physical inspection, and that visible conditions should not be the sole basis for adverse underwriting actions, with imagery generally required to be no more than 12 months old.  

The message from regulators across the US is consistent – the technology is legitimate but its application requires accountability. Scores must be explainable, adverse actions must be specific, and policyholders must have a meaningful opportunity to respond. 

Underwriting AI and the governance question 

That accountability obligation is now being systematically codified. The NAIC’s position is that AI is a tool used in underwriting, pricing, claims, and fraud detection, but it does not alter insurers’ legal obligations – existing state insurance laws apply regardless of whether decisions are made by humans, algorithms, or third-party vendors.  

The NAIC also recently launched a multistate pilot of its AI Systems Evaluation Tool across multiple states. The tool gives regulators a structured framework for reviewing insurer AI systems during market conduct examinations, covering AI governance structure and board oversight, model development and validation, data quality and bias testing, third-party vendor management, and documentation adequacy. 

The vendor question is particularly significant. Insurers are responsible for vendor compliance – that obligation does not transfer to the third party supplying the model. A carrier using an external AI platform for property risk scoring owns the governance requirements attached to every decision that platform generates. A model law on third-party data and models is anticipated in 2026, potentially including licensing requirements for vendors, with insurers expected to implement stricter contractual controls, documentation of model origins, and standards for explainability. 
 

“Technology helps in three ways. Granularity – enriching every submission with property-level intelligence so insurability rests on the specific risk, not a postcode average; speed – digitizing and triaging CAT-exposed submissions automatically so scarce expertise is reserved for the genuinely complex risks; and consistency – applying the same appetite and accumulation rules to every risk, every time, which matters enormously when managing aggregation in peak zones”
Richard HartleyCytora

Loss prevention as the new competitive ground  

The governance conundrum and the prevention opportunity are, in many ways, two sides of the same shift. As AI and real-time data move insurers closer to the individual property, the logical next step isn’t just better pricing – it’s intervening before the loss occurs at all. Most homeowners will go decades without filing a claim. AI enables insurers to support them in the years between losses through prevention, insight, and services that actively reduce risk. The shift from paying for risk to helping customers avoid it will define the next chapter of the industry. 

And that shift is already underway in how leading carriers are positioning themselves. Smart home devices that detect water leaks, monitor HVAC performance, and sense early fire indicators are shifting the insurer’s role from reactive claims payer to proactive risk partner. The commercial logic is straightforward here too – a leak sensor that shuts off a water supply before a pipe fails is worth more to both the insurer and the policyholder than any claims process, however efficiently run. Some insurers are integrating sensor data directly into loss control and claims processes, using early alerts to prevent or minimize damage, while others are analyzing anonymized sensor data to understand risk trends, refine underwriting, and improve claims efficiency.  

The data supporting loss prevention as a genuine performance lever is compelling. Research across property insurance segments shows that smart home technologies, including cameras, motion detectors, smart locks, and environmental sensors, demonstrably mitigate risks such as burglary, fire, and flooding, with multiple studies reporting fewer claims and reduced losses due to real-time monitoring.  

Competitive differentiation in the US property market will also hinge less on rate and more on risk governance, prevention partnerships, and resilience engineering. In a softening commercial property market, where rate alone is an increasingly blunt instrument, the carriers building prevention services into their proposition – monitoring programs, mitigation advice, sensor partnerships embedded in coverage – are creating a retention advantage that pricing cannot easily replicate. Clients are looking to insurers to go far beyond traditional coverage and act as true partners in risk consulting, with on-the-ground prevention capability increasingly critical as the frequency and severity of natural catastrophes continues to rise.  

[INSERT INFOGRAPHIC 2]

The property risk assessment of 2030: what gets built next 

So, what comes next? Well, the preceding themes in this guide – real-time intelligence, climate data reconstruction, AI-driven granularity, governance accountability, loss prevention partnerships – are not isolated trends; they're actually converging. And the logical destination of that convergence, four years from now, looks fundamentally different from the property risk assessment infrastructure that exists today. 

“Three trends stand out [this year],” adds Hartley. “First, the shift from reactive to predictive – the lesson from cyber, where insurers moved from pricing yesterday’s losses to anticipating tomorrow’s, is now spreading to property and CAT, where continuous, data-rich monitoring is replacing point-in-time assessment. 

“Second, the ecosystem itself has become the strategy: no single carrier owns all the data it needs, so the firms pulling ahead are those that orchestrate best-in-class third-party intelligence – catastrophe models, climate science, real estate and geospatial data – into a single, decision-ready view of risk.  

“Third, agentic AI is moving from experiment to operating model, with leading carriers building their underwriting around risk-data technology rather than bolting it on, decoupling premium growth from headcount. The throughline is that data and AI are becoming the core underwriting infrastructure rather than a side tool.” 

Manual underwriting will fundamentally change in the years to come, with certain tasks becoming obsolete, AI models automating the process and carriers offering real-time pricing based on dynamic risk assessment. In the meantime, the underwriter’s role may narrow toward complex risks where judgment still outperforms automation.  

For brokers, as pricing becomes data-determined and underwriting increasingly autonomous, the value proposition shifts entirely toward risk advisory – interpreting intelligence, identifying mitigation opportunities, and navigating a regulatory environment that will only grow more demanding. The annual renewal was once the primary touchpoint, but by 2030, it will be the least interesting part of the relationship.  

• 
  Cytora