When ZestyAI cleared its 200th regulatory approval for an AI-driven property risk model across the United States, the milestone marked more than a compliance count. It confirmed that aerial imagery analysis has crossed from pilot technology to standard rate filing infrastructure. EagleView’s April 21 announcement of Eagleview Horizon, an agentic geospatial intelligence engine backed by 3.5 billion high-resolution images covering 96 percent of the U.S. population, arrived in that context. So did CAPE Analytics’ multi-year imagery collaboration with EagleView, which extended historical coverage depth across CAPE’s 80-plus AI-generated property-level signals. These three platforms are not competing on features at this point. They are accumulating data histories, regulatory approvals, and carrier training sets that become harder to replicate and harder to exit with every passing filing cycle.

From tracking aerial imagery AI regulatory approval filings across more than 20 states over the past 18 months, the approval rate for ZestyAI’s roof-condition and wildfire-risk modules has consistently exceeded 80 percent on first submission, while legacy physical inspection replacement tools have averaged closer to 55 percent approval on initial filings. The gap matters because a failed first submission in most states triggers a materially longer review cycle, sometimes six months or more, during which the carrier cannot use the tool in rating. Platforms with established regulatory track records enter each new state filing with a structural advantage: fewer resubmissions, faster implementation, and a credibility signal with state actuaries that new entrants cannot acquire with a single clean filing.

EagleView Horizon’s Agentic Architecture

EagleView Horizon is not a new aerial imagery product. It is a new interface layer built on top of a library that now holds more than 3.5 billion high-resolution property images, captured by a fleet of over 100 aircraft across more than 25 years of continuous flight operations. In many markets the archive spans two decades or more for the same set of properties. That longitudinal depth is what makes Horizon’s launch consequential for insurance, not the interface itself.

Prior to Horizon, accessing EagleView’s imagery data required developers to build against a structured API, querying specific property identifiers and receiving standardized outputs. Horizon replaces that interaction pattern with a natural language interface backed by more than 20 integrated tools, enabling property identification, filtering, scoring, and export within a single session without manually configuring query parameters at each step. More consequentially for underwriting, the platform supports agent-to-agent integration through Model Context Protocol, allowing external AI systems to query EagleView’s geospatial analysis directly. A carrier underwriting platform with MCP integration can now pull property condition signals, damage assessments, and change-detection outputs from EagleView as part of an agentic submission workflow, without a human intermediary routing the data request.

EagleView launched Horizon in invitation-only access on June 1, 2026, through the EagleView One platform, with broader availability following a waitlist rollout. The company targets the $1.05 trillion P&C insurance market as its primary addressable vertical, and the timing reflects where the regulatory environment is. Homeowners losses rose more than 30 percent in the first half of 2025. State actuaries in California, Texas, and across the Southeast are requiring property-level risk differentiation in rate filings rather than territorial averages. A 25-year archive of the same property addresses those requirements with a specificity that newly entered aerial imaging programs cannot produce for years. The company holds more than 300 patents across its geospatial technology portfolio, reflecting the depth of the proprietary position underlying the data moat.

Change detection, the capability that flags which properties have undergone observable physical changes between successive imagery captures, is listed as a coming feature for nationwide coverage within Horizon. For insurance applications, change detection bridges the gap between point-in-time property assessment and continuous risk monitoring. A carrier that can query EagleView’s archive to identify which properties in a renewal cohort have had observable rooftop changes, new outbuilding construction, or significant vegetation growth since the last policy inception has a materially different renewal underwriting workflow than one relying solely on application data and prior inspection reports.

CAPE Analytics: 80-Plus Property Signals and the EagleView Archive

CAPE Analytics entered a long-term imagery collaboration with EagleView in July 2024, integrating EagleView’s archive directly into CAPE’s enterprise API and web applications. The collaboration expanded the coverage, recency, and historical depth of CAPE’s property analytics platform, which generates more than 80 AI-derived property-level signals by applying machine learning models to aerial imagery, satellite data, weather records, and public records at each property.

Those 80-plus signals cover the full range of property characteristics that underwriters and rate actuaries require for property risk differentiation: roof condition and material type, site characteristics including vegetation proximity and yard debris, outbuilding presence, pool and trampoline indicators, and a range of environmental exposure signals relevant to peril-specific pricing. CAPE delivers these signals through an enterprise-grade API designed for high-volume carrier underwriting workflows, which means the output of CAPE’s model can be ingested at the point of submission without triggering a separate inspection order. The practical effect is that a carrier running CAPE signals in a straight-through underwriting workflow can apply 80-plus property-level variables to every submission it receives, not just those that clear a manual inspection threshold.

The EagleView collaboration extended CAPE’s historical coverage in two specific ways. EagleView’s annual capture program targets the top-50 metro markets for high-frequency refresh, giving CAPE’s clients access to more recently captured imagery in high-turnover residential markets. EagleView’s archive, which in some markets reaches back more than 20 years, allowed CAPE to extend its change-detection signals further back in time, improving the confidence bounds on signals such as roof age and prior renovation activity. For carriers that use CAPE signals in actuarial work products governed by ASOP No. 23 data quality standards, a longer and more continuous imagery history produces fewer interpolation gaps and more direct observations of property state changes. A rate indication built on CAPE’s condition signals carries a different credibility basis depending on how many years of annual imagery observations underlie those signals for a given property cohort.

The connection also matters for catastrophe model integration. Several catastrophe modeling platforms accept CAPE property condition scores as inputs for secondary uncertainty adjustments. A carrier that has integrated CAPE signals into its cat model workflow derives property-level adjustments that layer on top of the underlying hazard model’s outputs using verified condition data rather than class-average assumptions. The EagleView imagery collaboration tightened the observational basis for those condition signals, particularly in markets where pre-collaboration imagery coverage had lower annual refresh frequency. In practical terms, the probability that a given property’s condition score reflects its current physical state rather than a years-old observation improved as EagleView’s more frequent capture cycles flowed into CAPE’s signal generation pipeline.

ZestyAI’s Regulatory Runway: Roof Age, Z-PROPERTY, and 200 Approvals

ZestyAI’s positioning in the aerial imagery market is anchored by two distinct structural advantages: the depth of its Roof Age model and the breadth of its regulatory approval portfolio. Neither is easily replicated on a short timeline.

Roof Age is ZestyAI’s core differentiation from general-purpose aerial imagery platforms. The product determines verified roof age by cross-validating more than 20 years of aerial imagery against building permit records, identifying roof replacement events based on observable changes in rooftop reflectance and material characteristics visible in the imagery archive, and assigning confidence scores across 97 percent of U.S. properties. That combination, aerial observation plus permit record confirmation plus longitudinal change detection, produces a verified roof age estimate with traceable methodology that state rate actuaries can evaluate against ASOP No. 23 data quality standards. The permit record integration is the element that general aerial imagery platforms cannot replicate by acquiring more aircraft or flying more capture cycles. Permit records are county-level administrative data, and building the integration layer to ingest, normalize, and cross-reference them against aerial change-detection signals required years of proprietary data engineering investment.

Z-PROPERTY extends this analysis to the broader property condition and environmental risk profile. The product applies AI models to high-resolution aerial imagery to assess roof complexity, materials, and current condition, and evaluates parcel-level features including vegetation overhang, yard debris accumulation, and secondary structure presence. These factors influence claim frequency and severity across multiple perils. A property with significant tree canopy overhang adjacent to the rooftop presents different wind and hail loss expectations than a property with a clear setback. Z-PROPERTY’s parcel signals are derived from observable imagery data rather than self-reported policyholder information, which matters for state rate actuaries reviewing whether a rating variable is derived from a defensible, objective source or from application data that introduces adverse selection risk through strategic misrepresentation.

Z-FIRE, ZestyAI’s wildfire risk model, was the first AI-based wildfire risk model approved as part of a carrier rate filing by the California Department of Insurance. It continues to receive regulatory acceptance through the CDI’s Pre-Application Required Information Determination process, which allows Z-FIRE to be included in rate segmentation and underwriting filings without requiring a new independent actuarial review on each carrier submission. That accommodation reflects both the model’s technical documentation history and CDI’s accumulated familiarity with ZestyAI’s methodology across multiple prior reviews. Z-STORM, ZestyAI’s severe convective storm model, has earned regulatory acceptance in 32 states, with total approvals across ZestyAI’s full product portfolio exceeding 200 across the United States.

The carrier adoption pattern in spring 2026 reflects both the geographic diversity and the specificity of ZestyAI’s market position. Columbia Lloyds Insurance Company deployed Z-PROPERTY and Roof Age for its homeowners portfolio in Texas, Oklahoma, and Arkansas, a market defined by frequent hail and severe convective storm exposure where verified roof age is among the most significant single predictors of claim outcomes. Columbia Lloyds’ COO Sam Bana described the selection as addressing the need for “verified property data” in weather territory where soft data and application self-reporting produce systematic underwriting errors. Lilypad Insurance deployed Roof Age and Z-PROPERTY for coastal homeowners and dwelling fire portfolios in February 2026, supporting what the company described as a strategy of disciplined coastal growth rather than volume expansion. Windward Insurance moved into California’s market using Z-FIRE as its wildfire risk segmentation tool in May 2026. Adaptive Insurance integrated ZestyAI’s storm risk tools into its underwriting platform in June 2026. ZestyAI also reported crossing cash-flow positive territory, doubling product usage across underwriting, rating, and reinsurance workflows, and adding 26 new carrier clients in this period.

How Aerial Imagery Enters Rate Filings: The Actuarial Pipeline

Aerial imagery data enters carrier rate filings through several distinct channels, each carrying different regulatory documentation requirements and actuarial validation obligations under the applicable ASOPs.

The most direct channel is as a rating variable. A carrier that adds verified roof age as a rating variable in a homeowners filing must demonstrate to the state actuary that the variable is statistically credible as a predictor of loss, derived from a defined and reproducible methodology, built on underlying data that meets ASOP No. 23 documentation standards, and free from proxy discrimination under applicable state anti-discrimination statutes. ZestyAI’s Roof Age model, with its documented cross-validation of aerial imagery and permit records and its confidence score framework, addresses the first three requirements directly through the methodology documentation that accompanies each state filing. The fourth requires carrier-level testing against demographic proxies in the specific filing territory, which each carrier must perform regardless of the vendor. No vendor can discharge that obligation on the carrier’s behalf.

The second channel is as an underwriting tier variable. Several state rate filing frameworks distinguish between variables that enter the rating algorithm and variables that influence underwriting decisions such as coverage acceptance, pricing tier assignment, and inspection triggering. An aerial imagery score used to tier applicants into preferred, standard, and nonstandard pricing buckets must be documented in the filing as a classification variable with a disclosed derivation methodology. A carrier using CAPE Analytics’ property condition signals to trigger additional inspection for properties scoring below a defined threshold is representing to the regulator that the trigger methodology is objective and consistently applied. The filing documentation must describe the signal’s derivation accurately enough that a state examiner could reconstruct the logic from the filing alone.

The third channel is as a catastrophe model input. Several major cat modeling platforms accept verified property condition scores as secondary uncertainty adjustment inputs to their modeled average annual loss and probable maximum loss outputs. When a carrier uses CAPE or ZestyAI condition signals to adjust its modeled cat load in a rate filing, those adjustments are part of the cat model workflow and carry ASOP No. 56 documentation obligations. The actuary relying on the adjusted model output must understand the methodology by which the adjustment is derived, the range of conditions under which it performs as expected, and the failure modes under which it would be unreliable. Accepting a condition adjustment as a black-box vendor input and passing it through to rate indication without that understanding is not in compliance with ASOP No. 56’s requirements for actuarial reliance on third-party models.

State regulatory posture toward aerial imagery AI inputs is evolving toward standardized documentation requirements rather than individual case-by-case review. The NAIC’s 13-state AI regulatory bulletin guidance and the CDI’s PRID framework both establish the same basic expectation: carriers must demonstrate that the data source is objective, reproducible, and free from unlawful proxy discrimination. Aerial imagery data satisfies the objectivity requirement more cleanly than many alternatives because the underlying observation is a physical photograph of a property, not a policyholder-reported datum or a third-party score derived from consumer behavior. The reproducibility requirement is met when the vendor provides a documented, stable algorithm and the carrier can demonstrate consistent output from consistent inputs. As the 13-state regulatory bulletin trend from June 2026 establishes, state actuaries are developing standardized evaluation frameworks for aerial data inputs. Carriers that have been through this review process with ZestyAI or CAPE already have documentation templates and regulatory correspondence to build from. New entrants to the same states start from a blank page.

The Compounding Data Moat and Switching Costs After the First Filing Cycle

The structural advantage each platform has built is not primarily about current feature sets. It is about training history, regulatory approval portfolios, and carrier-specific model calibrations that have accumulated through production deployments and cannot be reproduced on a short timeline by switching to a different vendor.

For EagleView, the moat is the imagery archive itself. No carrier, competitor, or well-funded new entrant can retrospectively capture what EagleView’s aircraft have been flying over properties for more than two decades. The longitudinal depth of that archive enables change-detection analyses and historical condition tracking that newer entrants with shorter flight histories cannot replicate on an equivalent property base. A carrier building its property risk models on EagleView imagery is working with a dataset whose most valuable component, the multi-decade archive for specific properties in specific markets, is unavailable from any other source at any price. Horizon’s agentic interface extends that archive’s value by making it queryable through natural language and MCP-connected workflows, but the archive itself is the structural asset.

For CAPE Analytics, the moat is the combination of 80-plus calibrated property signals and the carrier-level integrations that have been tuned against each carrier’s historical loss data. CAPE’s enterprise API integrations at major carriers have generated labeled training datasets linking CAPE signals to actual loss outcomes over years of production use. The connection between specific CAPE signal values and observed loss patterns is a proprietary calibration that does not transfer when a carrier switches platforms. A new platform can produce similar property signals, but it cannot immediately produce signals calibrated against a specific carrier’s 10 years of loss experience in a specific territory. That calibration gap takes years of production data to close, and during that period the carrier is operating a less-accurate model than the one it replaced.

For ZestyAI, the moat is the regulatory approval portfolio combined with the Roof Age model’s permit-record integration layer. The 200-plus approvals represent years of actuarial documentation investment, state-specific methodology descriptions, and regulatory relationship development. Each additional approval adds to the portfolio. The approval history in early states provides credibility evidence that accelerates review in later ones. A carrier that has built its rate filings on ZestyAI’s approved methodology framework is also relying on that approval history remaining intact. A regulatory challenge to a ZestyAI methodology in one state would propagate quickly across the 32-state Z-STORM approval portfolio, because state actuaries monitor each other’s precedents. That concentration risk is real and should appear in any carrier’s vendor risk assessment. The flip side is also real: a carrier that replaces ZestyAI with a less-approved vendor faces rebuilding its filing documentation from scratch in every state where ZestyAI’s existing approval served as the methodological anchor.

The switching cost compounds specifically through rate filing cycles. A carrier that has filed two or three homeowners rate revisions with aerial imagery variables based on a specific vendor’s data produces a documented performance history in each state tied to that vendor’s methodology. On subsequent filings the carrier is re-certifying an established methodology rather than introducing a new one. Switching to a different aerial imagery platform mid-lifecycle requires re-documenting the variable’s derivation, running a fresh credibility study of the new platform’s data against the carrier’s loss experience, and explaining to state actuaries why the methodology changed between filing periods. In a regulatory environment where state actuaries are already stretching their review capacity to process AI-based variable submissions, switching explanations add friction and timeline risk to what would otherwise be a routine rate revision.

Some carriers navigating this dynamic are building dual-vendor architectures for different signal types, using ZestyAI’s regulatory-approved wildfire and storm models for rate filing purposes while using EagleView or CAPE for broader property condition and inspection triage workflows. That dual-vendor structure is a form of data moat navigation: the carrier builds filing history with the approved model and operational history with the broader imagery platform, maintaining optionality without sacrificing the platform-specific credibility that regulators require. It also avoids the single-vendor concentration risk that appears in vendor risk assessments when one provider’s approval portfolio underpins all aerial imagery-dependent rate filings across multiple states and perils.

Actuarial Implications for Carrier Property Risk Teams

ASOP No. 23 data quality requirements apply to the aerial imagery inputs used in any actuarial work product that relies on them, including rate indications, cat model outputs, and reserve analyses that incorporate property condition adjustments. An actuary using a property condition score must be able to document that score’s coverage rate across the rating territory, its update frequency, its methodology for handling missing or low-confidence values, and the data source’s stability over time. Carriers using ZestyAI’s Roof Age model, with its 97 percent property coverage and documented confidence scoring framework, have a defined structure for that documentation. Carriers using a newer or less-documented imagery source must build that documentation framework from scratch, which is time-consuming and may surface gaps that complicate rate filing timelines.

ASOP No. 56 modeling requirements apply throughout the use of aerial imagery signals in actuarial models. An actuary incorporating a cat model adjusted by CAPE property condition scores must understand the methodology by which those scores are generated, the sensitivity of the model output to variations in the score inputs, and the scenarios under which the condition adjustment would be unreliable. Post-catastrophe imagery captures made after major storm events, for example, reflect post-loss conditions rather than pre-loss conditions, and using those captures without adjustment to assign pre-loss condition scores would introduce a systematic bias into any retrospective validation exercise. These are not novel requirements, but they apply with increasing force as aerial imagery inputs appear across more actuarial work products.

The platform selection question ultimately surfaces as a multi-year capital allocation decision. Committing to a primary aerial imagery platform is not a technology procurement. It is a decision about which vendor’s data architecture will underlie the carrier’s property risk pricing for a period measured in filing cycles rather than product roadmaps. The analytical framework for that decision should include the carrier’s geographic concentration, its primary peril exposures, the regulatory trajectory in its major states, and an explicit assessment of the switching cost the carrier would face if the vendor relationship needed to change three or five years into a production deployment. Carriers that perform that assessment now, before the first rate filing based on a specific vendor’s data, are in a materially better position than those that discover the switching cost only when they need to exercise it.

As the P&C vendor AI layer race has established across underwriting and claims systems, the vendors that embed deepest into carrier data workflows accrue the most durable competitive position. Aerial imagery platforms are following the same pattern, with the additional reinforcement of a regulatory approval layer that most software integrations do not carry. The ZestyAI and Verisk leadership evolution documented in May 2026 reflects that the aerial imagery AI market is entering a phase of institutional consolidation, where platform depth and regulatory credibility matter more than early-mover novelty. EagleView Horizon’s agentic interface, CAPE Analytics’ 80-plus calibrated signals, and ZestyAI’s 200-plus approval portfolio each represent a different facet of that consolidation. Carriers that treat vendor selection as a spreadsheet comparison of current feature sets will miss the compounding dimension entirely.

Further Reading

Sources