LexisNexis Scores Home Risk Across Six Perils: The Validation Problem Actuaries Should Ask First

LexisNexis Risk Solutions launched Location Intelligence for Home on May 14, 2026, an AI model that scores homeowners properties across six perils and claims a 20-fold gap in claim frequency between its top- and bottom-scoring properties (LexisNexis Risk Solutions, May 2026). For an actuary, that lift is a hypothesis to be tested by peril, territory, and cohort, not a number to be accepted at face value before it reaches a rate filing.

A Model Built for the Peril Underwriting Tools See Least Well

Location Intelligence for Home is embedded in Smart Selection, LexisNexis's existing automated data service that issues inspection flags and configurable business rules for new business and renewal underwriting (LexisNexis Risk Solutions, May 2026). The model uses neural network techniques trained on industry-wide home claims data, location-based signals, and historical loss patterns to generate a single property-level score across six perils: hail, wind, weather-related water, non-weather-related water, freeze, and collapse or falling object. "Rising loss costs and shifting risk patterns are making it harder for home insurers to rely on traditional underwriting approaches alone," said George Hosfield, vice president of home insurance at LexisNexis Risk Solutions (LexisNexis Risk Solutions, May 2026). The product follows a similar rollout LexisNexis already completed in commercial lines, and the company says it will file Location Intelligence for Home as a predictive model in multiple states in the coming months for use in underwriting and rating workflows.

The six-peril bundling is the design choice that matters most to an actuary, because it merges two very different risk families into one score. Hail and wind sit adjacent to catastrophe modeling, where exterior condition, roof age, and geocoded storm exposure already drive most carriers' existing tools, including exterior-imagery vendors like ZestyAI, EagleView, and Moody's Cape Analytics that this site has covered building competing property-risk platforms on aerial and satellite data (see our analysis of the aerial-imagery data moat). Freeze, collapse, and above all non-weather water are different: they are attritional, plumbing- and structure-driven perils with no visible exterior signature, and they are exactly what a roof-condition score or a hail-swath overlay cannot see. LexisNexis's own numbers make the gap explicit. Non-weather water represented 24% of all home insurance claims in 2025, while weather-related water accounted for only 4% (LexisNexis Risk Solutions, May 2026). A peril responsible for roughly a quarter of the industry's claim volume has historically been underweighted by underwriting tools built around what an inspector or an aerial image can actually observe, which is precisely the segmentation gap Location Intelligence for Home is positioned to close, if the lift survives scrutiny peril by peril rather than only in the blended headline number.

Why a Blended 20x Lift Is Not the Number That Matters

A single combined lift statistic across six structurally different perils is close to useless for pricing purposes on its own, and it is worth being explicit about why. Claim frequency, expected severity, and loss-cost volatility differ enormously across hail, wind, weather water, non-weather water, freeze, and collapse. A model that separates properties well on hail exposure but poorly on non-weather water could still produce an impressive blended 20x figure if hail happens to dominate the score's variance, while doing almost nothing to solve the plumbing-deterioration problem the product was ostensibly built to address. The reverse is equally possible: strong separation on the attritional perils, masked by noisy performance on catastrophe perils where the score adds little over what carriers already model well.

This is the same pitfall that shows up whenever a vendor reports an aggregate lift metric for a multi-peril or multi-line score, and it is the reason the NAIC's own model-governance guidance for insurers, echoed in state model laws following the template of the NAIC's AI risk-management taxonomy, pushes carriers toward peril-specific and use-specific validation rather than accepting a single headline statistic. An actuary evaluating Location Intelligence for Home, or any comparable third-party score, should insist on six separate lift curves, one per peril, each built on a holdout sample the vendor did not use to fit the model. Anything less lets a single dominant peril carry the marketing number while the perils that matter most for loss-cost accuracy, freeze and non-weather water in particular, go unverified.

A Practical Peril-Level Validation Table

The Non-Weather Water Problem: What the Score Would Have to Prove

Non-weather water claims, principally failed supply lines, aging water heaters, undetected slab leaks, and appliance-hose failures, are the actuarial case where an AI location score has the clearest theoretical edge and the hardest validation burden. Traditional underwriting proxies for this risk indirectly: home age as a proxy for plumbing material and age, prior-loss history as a lagging indicator, and inspection flags that catch visible deterioration but miss anything behind a wall or under a slab. A model trained on industry-wide claims data can, in principle, learn patterns that home age alone cannot: neighborhood-level plumbing vintage correlated with subdivision-era construction codes, water-pressure and soil-condition interactions with pipe corrosion, or appliance-failure clustering tied to specific manufacturer cohorts sold in a given region and decade.

The actuarial test of whether that theoretical edge is real is straightforward to state and hard to skip: does the non-weather water lift survive after the carrier's own rating plan already controls for home age, a plumbing-material proxy where available, and prior-loss count? If the vendor's peril-specific lift collapses once those variables are held fixed, the score is not adding new information; it is re-deriving what the carrier's existing variables already capture, dressed in a more impressive-looking number. If the lift persists, the carrier has a legitimate case for a new rating or underwriting variable, and the next question becomes whether that variable belongs in the rating plan itself or in an underwriting referral rule, which is a different regulatory and operational decision.

Underwriting Referral Versus Rate Classification: Two Different Uses, Two Different Bars

Smart Selection's stated function, inspection flags and configurable business rules, points toward underwriting referral as the near-term use case: a high Location Intelligence score triggers a required inspection, a coverage restriction, or a non-renewal review, rather than directly setting the premium. That distinction carries real regulatory and actuarial weight, and conflating the two uses is a common vendor-adoption mistake.

A referral rule needs only directional validity and operational consistency: the score needs to reliably flag the properties worth a second look, and the carrier needs underwriters applying that flag the same way across territories and tenure, not inconsistently by branch office or individual underwriter judgment. Rate classification is a much higher bar. Filing a variable into a rating plan requires demonstrating actuarial soundness under the applicable state rating law, typically meaning the variable is predictive, not unfairly discriminatory, and stable enough that policyholders are not repriced on noise. A score that clears the referral bar comfortably, useful signal, applied consistently, can still fail the classification bar if its lift is unstable across renewal cohorts, correlates too tightly with protected-class proxies at the ZIP level, or has not been demonstrated with the same actuarial rigor a carrier would apply to an in-house generalized linear model or gradient-boosted rating variable. The practical sequencing most carriers should follow: deploy a new third-party score in underwriting referral first, accumulate at least a full renewal cycle of its own experience data, and only then evaluate whether the peril-specific, cohort-stable lift supports a rate filing.

The Regulatory Backdrop: A Data Call That Will Make These Scores Auditable

The timing intersects directly with a separate NAIC initiative. State insurance regulators issued a nationwide homeowners market data call in spring 2026, covering policy years 2018 through 2025 and requiring insurers writing at least $50,000 in relevant premium to report peril-level claims, premiums, deductibles, cancellations, and non-renewals down to the ZIP-code level (NAIC, 2026). The original submission deadline of June 15, 2026, drew a request for a one-month extension to July 15, 2026, and the NAIC has said it plans to release a public report on the aggregated findings in early 2027 (NAIC, 2026). That data call was not built with any single vendor's model in mind, but it will generate, for the first time, a peril-level, ZIP-level public dataset against which independent researchers and regulators can sanity-check whether AI-driven underwriting tools are producing the segmentation gains vendors claim, or whether they are simply reallocating coverage away from ZIP codes correlated with income or demographic characteristics. Any carrier adopting a third-party score like Location Intelligence for Home should expect that its territory-level underwriting actions, non-renewal rates, referral triggers, coverage restrictions, become comparable against that public data once the 2027 report lands, whether or not the carrier itself intended that comparison.

That regulatory exposure compounds the case for the cohort-stability and drift checks in the validation table above. A model that produces geographically clustered non-renewal patterns will be visible in the NAIC's own ZIP-level data within a reporting cycle or two, and a carrier that cannot show its own peril-specific validation work for the score driving those patterns will have a harder time explaining the pattern to a market conduct examiner than one that can point to a full documented validation file.

Why This Matters

The actuarial stakes here are not about whether LexisNexis's model works, they are about what an insurer can prove before relying on it. A 20x blended lift claim is a starting hypothesis, not a finding, and the underwriting decisions built on top of it, referrals, non-renewals, eventually rate classification, need to rest on peril-specific, cohort-stable, drift-tested evidence the carrier has generated or independently reviewed, not on the vendor's launch marketing. Carriers that build a documented validation record now, peril-level lift curves, actual-to-expected ratios, renewal-cohort stability, and a clear referral-versus-classification sequencing plan, will be positioned to defend both their rate filings and their non-renewal patterns once the NAIC's 2027 homeowners data becomes the yardstick every home insurer's territory-level decisions get measured against.