Claim 1 of US Patent 12,700,042 does not score a roof once. It scores the same property twice, across a defined time window, and moves money on the difference: the method extracts feature vectors for discoloration, tarp presence, material degradation, missing material, sealing and defects from imagery, computes a condition score at each end of the window, then increases or decreases the premium based on the detected change (USPTO, granted August 4, 2026). A maintenance signal has become a rating trigger.

Key Takeaways

  • The second Cape-authored patent to publish within eight days, both now Moody's intellectual property after the January 2025 acquisition. The July grant decided what a property looks like; this one decides what to charge when that appearance changes.
  • The claim is silent on cadence. It requires two or more reads across a time window and prices the delta, but nothing ties that window to a policy's annual term, so it could run monthly or mid-term against a policy already in force.
  • At least 20 states have flagged discoloration as insufficient standalone grounds for adverse action in aerial-imagery underwriting bulletins. It is one of exactly six attributes the claim requires the model to extract.
  • More than 100 carriers already license Cape's static Roof Condition Rating, approved in more than 40 states across more than 300 filed rating and underwriting uses. A change-detection variant is the natural upsell.

Patent Details

Patent Number U.S. 12,700,042 B2
Filed January 29, 2024
Granted August 4, 2026 (Official Gazette week 31)
Assignee Cape Analytics (acquired by Moody’s, January 2025)
Classification CPC G06Q 40/08 (insurance business methods)
Named attributes Discoloration, tarp presence, material degradation, missing material, sealing, defects
Sibling patent U.S. 12,694,669 B1, granted July 28, 2026 under G06V 20/176 (image recognition)

What the Claim Actually Locks In

The independent claim is specific about mechanism rather than outcome. It determines measurements depicting a property including imagery across a time window, derives a semantic segmentation mask for each, isolates property-feature pixels and extracts a feature value vector from only those pixels. It then applies a per-feature scaling factor, runs a machine learning model to turn the scaled vectors into attribute values, produces a condition score at each end of the window, and increases or decreases the premium based on the change and the score itself.

The claim enumerates the semantic attributes by name: discoloration, tarp presence, material degradation, missing material, sealing, or defects. That is a meaningfully different object than Cape's existing Roof Condition Rating, which is already approved for ratemaking in more than 40 states and is, like most vendor condition scores, a point-in-time read refreshed at renewal from a single current image.

The reason a vendor patents a change detector rather than resting on a static score is that homeowners losses are disproportionately roof losses. Wind and hail damage accounted for the largest share of homeowners claims from 2018 through 2022, with 2.8% of insured homes filing a wind or hail claim in that window, roughly one in 35 in any given year (Triple-I).

Roof age and condition are already standard rating variables. What they do not capture is trajectory. A 12-year-old roof photographed in good condition at last year's renewal and a 12-year-old roof photographed deteriorating today carry the same age band and, on many current rating plans, the same relativity. The condition-change score is built to close that gap, and the silence on cadence is what makes it more than a better renewal variable.

Discoloration Is Exactly What Regulators Already Flagged

Because the claim is silent on cadence, it raises a ratemaking question a renewal-time variable never faces: does a detected change move premium prospectively at renewal, where a filed relativity change is disclosed on the renewal notice, or mid-term against a policy already in force and an exposure basis already earned? States regulate those paths very differently. New Jersey bars mid-term premium increases and coverage reductions absent prior written approval from the Commissioner, and other states echo it.

The sharper problem sits in the attribute list. Connecticut issued a bulletin in March 2024 after consumer complaints about nonrenewals based on aerial imagery. Under state law, mere discoloration, streaking, or other cosmetic issues that do not affect structural integrity are not valid grounds for nonrenewal, and if imagery does not definitively show material damage the carrier must obtain a physical inspection first. By July 2026, at least 20 states had adopted comparable bulletins.

A rate filing and a nonrenewal notice are legally distinct instruments under separate statutes. But a premium increase driven by a detected discoloration score functions, from a policyholder's vantage point, almost identically to the practice those bulletins were written to stop: an adverse financial consequence triggered by a cosmetic imagery signal regulators have already said cannot stand alone.

Underneath sits a second exposure. The NAIC's AI model bulletin, adopted in 25 states as of July 2026, requires documented testing of AI-driven outputs including third-party vendor scores for unfair discrimination against protected classes. Deferred-maintenance signals correlate with household income and, in some markets, with historically disinvested neighborhoods where deferred maintenance tracks financial constraint rather than risk-taking.

A carrier filing this score therefore needs disparate-impact testing run against the change-detection output specifically, not inherited from whatever it already ran on the static variable. A delta built from six cosmetic and structural features can encode a different demographic signature than a single point-in-time value. The lower-friction path is to route the signal into loss-control outreach instead, a free inspection or a maintenance credit, which sidesteps the bulletins entirely because no adverse pricing action occurs.

What Moody's Now Owns

Moody's announced the Cape acquisition on January 13, 2025, terms undisclosed, folding the geospatial property intelligence into its catastrophe models (Moody's, January 2025). Both patents run on the same aerial-imagery pipeline: one governs how disagreeing attribute readings reconcile into a confident value, the other how a change in that value gets priced.

Together they extend the pattern the site has tracked across EagleView, Cape, and Zesty.ai: the durable advantage is not the image-recognition model, now table stakes, but the proprietary logic layered on top. What likely got the claim past Section 101 is the same specificity, since it recites segmentation masks, pixel isolation, per-feature scaling and a named six-item attribute list rather than claiming "price insurance using AI," matching the Section 101 reset the site found across other recent grants.

The complication is who carries the validation burden. State Farm has pursued the same continuously-updated idea from IoT sensors on commercial property, but those claims run on first-party data the carrier controls. Cape's run on externally licensed imagery a carrier buys as a vendor feed, so the carrier files and defends a score built inside a model it does not own and cannot fully reproduce.

That leaves the more than 100 carriers already licensing the static rating most exposed, because the upsell arrives as a small extension to a variable they have already filed. The grant gives Moody's exclusive rights to a specific method for turning a sequence of property photographs into a premium decision. Whether a given state treats that decision as a defensible filed rate factor, or as the same aerial-imagery problem it spent 2024 through 2026 writing bulletins against, is a filing-by-filing question the patent itself leaves open.

Further Reading

Sources