DCT

1:25-cv-00201

Aon Re Inc v. Zestyai Inc

Log In

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 1:25-cv-00201, D. Del., 02/19/2025
  • Venue Allegations: Venue is alleged to be proper in the District of Delaware because the Defendant is incorporated in Delaware.
  • Core Dispute: Plaintiff alleges that Defendant’s AI-driven property risk analysis platform infringes four patents related to using aerial imagery and machine learning to automatically assess property characteristics and conditions for insurance underwriting.
  • Technical Context: The technology operates in the "insurtech" sector, automating property risk assessment for the insurance industry, a task that has traditionally involved manual inspection and less sophisticated data analysis.
  • Key Procedural History: The complaint alleges that Defendant had actual knowledge of U.S. Patent Nos. 10,529,029 and 10,650,285 since the summer of 2020. It further alleges that Plaintiff provided Defendant with formal notice of infringement of all patents-in-suit via a letter dated April 20, 2023, followed by claim charts in September 2023. These allegations form the basis for the claim of willful infringement.

Case Timeline

Date Event
2016-09-23 Priority Date for '029, '285, '491, '058 Patents
2020-01-07 U.S. Patent No. 10,529,029 Issues
2020-05-12 U.S. Patent No. 10,650,285 Issues
2020-06-01 Alleged Actual Knowledge of '029 and '285 Patents by Defendant
2021-06-08 U.S. Patent No. 11,030,491 Issues
2021-12-07 U.S. Patent No. 11,195,058 Issues
2023-04-20 Plaintiff Sends Formal Notice Letter to Defendant
2023-09-01 Plaintiff Provides Claim Charts to Defendant
2025-02-19 Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 10,529,029 - "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Maintenance through Aerial Imagery Analysis"

  • Patent Identification: U.S. Patent No. 10,529,029, "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Maintenance through Aerial Imagery Analysis," issued January 7, 2020 (Compl. ¶10).

The Invention Explained

  • Problem Addressed: The patent's background section describes the challenge of populating insurance risk exposure databases, noting that traditional methods are inefficient. The inventors identified a need for technology that could "automatically extract characteristics of individual properties, providing fast and efficient automated classification of building styles and repair conditions" from aerial imagery ('029 Patent, col. 2:6-11).
  • The Patented Solution: The invention proposes a computer-implemented method that obtains an aerial image of a property and extracts features from it, described as "pixel groupings." The core of the solution is a two-stage analysis using distinct machine learning classifiers: a first classifier determines a "property characteristic classification" (e.g., the shape of a roof), and a second classifier determines a "condition classification" (e.g., the state of repair of that roof) ('029 Patent, col. 24:17-48). This dual classification is then used to determine a risk estimate. The system architecture is depicted in Figure 3, which illustrates separate engines for characteristic classification (322) and condition classification (324) ('029 Patent, Fig. 3).
  • Technical Importance: This automated, multi-classifier approach was developed to increase the speed, efficiency, and accuracy of property risk assessment for insurance underwriting (Compl. ¶¶18, 28).

Key Claims at a Glance

  • The complaint explicitly quotes and alleges infringement of independent claim 15, a non-transitory computer-readable medium claim (Compl. ¶50). It also asserts claims 1, 2, 3, 5, 18, 19, and 20 (Compl. ¶49).
  • Essential elements of independent claim 15 include instructions for a computer to:
    • Receive a request for property classification from a user.
    • Obtain an aerial image of the property from a remote source.
    • Extract features from the image, where the features include "pixel groupings."
    • Apply the pixel groupings to a first machine learning classifier to determine a property characteristic classification.
    • Apply the pixel groupings to a second machine learning classifier to determine a property condition classification.
    • Determine a risk estimate based on the classifications.
  • The complaint reserves the right to assert additional claims (Compl. ¶71).

U.S. Patent No. 10,650,285 - "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Conditions through Aerial Imagery Analysis"

  • Patent Identification: U.S. Patent No. 10,650,285, "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Conditions through Aerial Imagery Analysis," issued May 12, 2020 (Compl. ¶11).

The Invention Explained

  • Problem Addressed: The patent addresses the same technical problem as the ’029 Patent: the need for a more efficient and automated system for analyzing property features from imagery for risk assessment in the insurance industry (’285 Patent, col. 1:18-24).
  • The Patented Solution: The invention describes a system comprising two distinct, pre-trained machine learning models stored in memory: a first model for identifying property characteristics and a second for identifying property conditions. The system extracts "image-related features" from an aerial image and applies a "first portion" of these features to the first model and a "second portion" to the second model to determine the respective classifications (’285 Patent, col. 34:11-30). This claimed division of features into "portions" for different models is a specific architectural element.
  • Technical Importance: The claimed multi-step architecture is presented as an improvement that enhances the speed, efficiency, and accuracy of automated property risk assessment (Compl. ¶¶29, 31).

Key Claims at a Glance

  • The complaint explicitly quotes and alleges infringement of independent claim 9, a system claim (Compl. ¶80). It also asserts claims 1-4, 6-10, 13, 15, 16, 17, 19, and 21 (Compl. ¶79).
  • Essential elements of independent claim 9 include:
    • A storage region storing a first machine learning analysis model (for characteristics) and a second machine learning analysis model (for conditions).
    • Processing circuitry with instructions to:
    • Extract a set of image-related features from an aerial image.
    • Apply a "first portion" of the features to the first model to determine a characteristic classification.
    • Apply a "second portion" of the features to the second model to determine a condition classification.
  • The complaint reserves the right to assert additional claims (Compl. ¶100).

Multi-Patent Capsule: U.S. Patent No. 11,030,491

  • Patent Identification: U.S. Patent No. 11,030,491, "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Conditions through Imagery Analysis," issued June 8, 2021 (Compl. ¶12).
  • Technology Synopsis: This patent describes a system that accesses a plurality of images of a property, applies boundary information to isolate the property within the images, and then uses a set of machine learning algorithms to determine property characteristics and a separate algorithm to classify the property's condition (Compl. ¶¶32-33, 109). The use of boundary information to isolate the property is a specific focus.
  • Asserted Claims: The complaint asserts independent claim 1 and dependent claims 2-3, 6, 8, 10-13, and 17-20 (Compl. ¶¶108-109).
  • Accused Features: The Z-PROPERTY platform is accused of infringing by accessing multiple images, using parcel data as "boundary information" to isolate properties, and then running its machine learning models to classify roof characteristics and conditions (Compl. ¶¶113-117).

Multi-Patent Capsule: U.S. Patent No. 11,195,058

  • Patent Identification: U.S. Patent No. 11,195,058, "Platform, Systems, And Methods for Identifying Property Characteristics and Property Feature Conditions through Aerial Imagery Analysis," issued December 7, 2021 (Compl. ¶13).
  • Technology Synopsis: This patent claims a system having a storage medium with pre-developed "property characteristic profiles" and "property condition profiles" created through machine learning training. In response to a user request, the system accesses aerial imagery and applies these stored profiles to the image to classify both the characteristics and condition of a property feature, then presents the information in a graphical user interface (Compl. ¶¶34-35, 136).
  • Asserted Claims: The complaint asserts independent claim 1 and dependent claims 2, 4-7, 10, 11, 13, 15, and 18 (Compl. ¶¶135-136).
  • Accused Features: The Z-PROPERTY platform is accused of infringing by storing and applying its pre-trained machine learning models (the "profiles") for characteristics (e.g., roof shape) and conditions (e.g., roof wear) to aerial imagery in response to user requests (Compl. ¶¶138-146).

III. The Accused Instrumentality

Product Identification

  • The accused instrumentality is Defendant's commercial offering marketed as "Z-PROPERTY," which is also referred to as "Location Insights" (Compl. ¶51). The platform includes specific modules such as "Digital Roof," "Z-VIEW," "Z-STORM," and "Z-HAIL" (Compl. ¶51).

Functionality and Market Context

  • Z-PROPERTY is described as a platform that provides "property risk and value intelligence to insurance and real estate through AI and fresh aerial imagery" (Compl. ¶53). The complaint includes a marketing image from Defendant’s website stating this purpose (Compl. ¶53).
  • Its technical function is to automatically analyze property information for use in insurance underwriting, risk selection, and rating (Compl. ¶52). The platform allegedly leverages "computer vision and deep learning on 115+Bn data points... to extract key building features" (Compl. ¶57).
  • Specific accused functionalities include using machine learning classifiers to identify roof characteristics, such as shape (e.g., "hip, gable, mixed, flat"), and roof condition, which is categorized into five levels from "Good" to "Major Damage" (Compl. ¶¶59, 60, 89, 90). The platform provides results to users via a graphical interface called "Z-VIEW" (Compl. ¶55).
  • The complaint alleges that Defendant uses the Z-PROPERTY platform to directly compete with Plaintiff in the insurance analytics market (Compl. ¶16).

IV. Analysis of Infringement Allegations

10,529,029 Infringement Allegations

Claim Element (from Independent Claim 15) Alleged Infringing Functionality Complaint Citation Patent Citation
receive, from a user at a remote computing device, a property condition classification request... Defendant's Z-PROPERTY platform, via its Z-VIEW application, allows a user to input an address or portfolio to access property insights, which constitutes receiving a request. ¶55 col. 3:25-30
obtain, from a remote data source responsive to receiving the property classification request, an aerial image of a geographic region including the property The Z-PROPERTY platform obtains and utilizes the "latest aerial imagery" covering North America to perform its analysis. ¶56 col. 3:31-34
extract from one or more of a plurality of features from the aerial image... wherein the extracted features include pixel groupings representing the respective property characteristic Defendant's system uses computer vision and deep learning to "extract key building features," and Defendant's own patent application allegedly discusses using "pixel groupings." ¶57, ¶58 col. 8:5-8
determine... a respective property characteristic classification... applying the pixel groupings for the respective property characteristic to a first machine learning classifier trained to identify property characteristics... The Z-PROPERTY "Digital Roof" feature uses a machine learning classifier to identify a roof's type or shape (e.g., hip, gable, mixed). ¶59 col. 3:40-44
determine... a respective condition classification... applying the respective pixel groupings for the respective property characteristic to a second machine learning classifier trained to identify property characteristic conditions... The Z-PROPERTY "Digital Roof" feature uses a machine learning classifier to categorize a roof's condition into one of five levels (e.g., "Heavy Wear"). A screenshot from Defendant's website shows this five-level condition categorization (Compl. ¶60). ¶60 col. 15:13-17
determine, in real-time responsive to receiving the property classification request... at least one risk estimate representing risk of damage due to disaster The Z-PROPERTY "Z-HAIL" feature provides a user with a "claim frequency and claim severity score that represent a risk estimate of damage." A screenshot of the Z-HAIL interface displays these risk scores (Compl. ¶61). ¶61 col. 4:6-11

Identified Points of Contention

  • Scope Questions: A central question may be whether Defendant's use of "artificial intelligence" and "deep learning" (Compl. ¶54) embodies the specific two-step architecture of a "first machine learning classifier" for characteristics and a separate "second machine learning classifier" for conditions, as recited in the claim. The analysis will likely focus on whether the accused system maintains this structural and functional separation of classifiers or uses a different architecture, such as a single, multi-output model.
  • Technical Questions: The complaint's allegation that Z-PROPERTY uses "pixel groupings" relies in part on Defendant's separate patent application (Compl. ¶58). A point of contention may be what technical evidence demonstrates that the features actually extracted and used by the accused Z-PROPERTY product are "pixel groupings" as that term is understood in the patent, rather than a different form of feature representation.

10,650,285 Infringement Allegations

Claim Element (from Independent Claim 9) Alleged Infringing Functionality Complaint Citation Patent Citation
a non-transitory computer readable storage region storing a first machine learning analysis model trained to identify one or more property characteristics, and a second machine learning analysis model trained to identify one or more property conditions The Z-PROPERTY system is alleged to comprise stored machine learning models, including a first model for identifying property characteristics (e.g., roof type) and a second model for identifying property conditions (e.g., roof wear level). ¶85, ¶86 col. 15:3-31
processing circuitry; and a non-transitory computer readable medium having instructions stored thereon Defendant's system operates on processing circuitry using "artificial intelligence to account for all factors that may impact a property's value and its risk exposure." ¶87 col. 21:24-32
extract a set of image-related features from an aerial image... Defendant's press releases state that its system uses "computer vision and deep learning... to extract key building features" from property data. ¶88 col. 7:61-8:1
apply a first portion of the set of image-related features to the first machine learning analysis model to determine a characteristic classification... The "Digital Roof" functionality allegedly applies features from an image to a machine learning model to determine the roof type (e.g., hip, gable, mixed). ¶89 col. 15:3-12
apply a second portion of the image-related features to the second machine learning analysis model to determine a condition classification... The "Digital Roof" functionality allegedly applies features from an image to a machine learning model to categorize the roof's condition into one of five levels. ¶90 col. 15:13-17

Identified Points of Contention

  • Scope Questions: The infringement analysis for this patent may turn on the construction of applying a "first portion" and a "second portion" of the image-related features to different models. The dispute may center on whether this requires distinct, segregated subsets of features being routed to separate models, or if it can be read more broadly to cover an architecture where different aspects of a single, unified feature set are analyzed by different models or parts of a model.
  • Technical Questions: A key evidentiary question will be whether Plaintiff can demonstrate that the internal architecture of Z-PROPERTY actually segregates and applies feature sets as "first" and "second" portions. The complaint's allegations are based on the product's externally-observed functions (classifying shape, classifying condition), which may not be sufficient to prove the specific internal data-flow architecture required by the claim.

V. Key Claim Terms for Construction

The Term: "pixel groupings" ('029 Patent, claim 15)

Context and Importance

  • This term defines the input data extracted from the aerial image that is fed into the two machine learning classifiers. Its construction is critical because it determines whether modern deep learning feature extraction techniques, which may produce abstract feature vectors rather than direct groupings of pixels, fall within the scope of the claim.

Intrinsic Evidence for Interpretation

  • Evidence for a Broader Interpretation: The claim language itself is general, reciting "pixel groupings representing the respective property characteristic," which could be argued to encompass any feature set derived from the image's pixels.
  • Evidence for a Narrower Interpretation: The specification provides examples of extracted features such as "angles, outlines, substantially homogenous fields" ('029 Patent, col. 8:5-8). A defendant may argue that these examples limit the term to more traditional, engineered features rather than the abstract, learned features of a deep neural network.

The Term: "first portion" / "second portion" [of the set of image-related features] ('285 Patent, claim 9)

Context and Importance

  • These terms describe how the extracted image features are divided and applied to the two separate machine learning models for characteristic and condition analysis. Practitioners may focus on this term because the infringement question may depend on whether Z-PROPERTY's internal data architecture involves such a division of its feature set.

Intrinsic Evidence for Interpretation

  • Evidence for a Broader Interpretation: A plaintiff may argue that the terms do not require physically separate or non-overlapping data sets. Instead, it could be interpreted to mean that the first model relies on one "portion" (or aspect) of a feature vector while the second model relies on another "portion" (or aspect) of the same vector.
  • Evidence for a Narrower Interpretation: A defendant may argue that the plain language implies a division of the "set" of features into at least two distinct subsets, which are then separately routed to their respective models. This would suggest a specific data flow that may not be present in an integrated AI system.

VI. Other Allegations

  • Indirect Infringement: The complaint alleges inducement of infringement for all four patents. The allegations are based on Defendant's promotional literature, website marketing, customer "Case Studies," and technical support, which allegedly instruct and encourage end-users to operate the Z-PROPERTY platform in an infringing manner (Compl. ¶¶74-75, 104, 131, 161).
  • Willful Infringement: Willfulness is alleged for all four patents based on both pre-suit and post-suit knowledge. The complaint alleges Defendant had actual knowledge of the '029 and '285 patents since "summer of 2020" and was put on formal notice of all four patents-in-suit by a letter dated April 20, 2023 (Compl. ¶¶14, 62, 91, 118, 148).

VII. Analyst’s Conclusion: Key Questions for the Case

  • A central issue will be one of architectural equivalence: does the accused Z-PROPERTY platform, described as using a general "AI" and "deep learning" engine, implement the specific two-classifier architecture (one for "characteristics," one for "conditions") that is a core limitation of the asserted patents, or is there a fundamental mismatch in its technical design?
  • The case may also turn on a question of definitional scope: can claim terms rooted in earlier machine learning concepts, such as "pixel groupings" and applying distinct "portions" of features, be construed broadly enough to read on the integrated and abstract feature extraction methods used in modern deep learning systems?
  • A key evidentiary question will be whether the public-facing functionality and marketing claims of the Z-PROPERTY platform provide a sufficient basis to plausibly infer that its undisclosed internal workings meet the specific structural and methodological limitations recited in the patent claims.