Allstate Patents Turn Road Risk Into Rating Evidence

On May 19, 2026, the USPTO issued Allstate Insurance Company’s patent US12632903, titled “Integrated ratemaking platform.” Fourteen days later, the Office issued US12644716, “Three-dimensional risk maps.” The two-week gap matters because the patents are not independent features. One captures and structures road-level risk evidence from vehicle sensors. The other connects analytics outputs to the production rating engine. Read together, they describe the completion of a data path Allstate has been assembling across at least three prior grants in the risk-mapping family: the gap between sensor observation and rate determination has a new architectural bridge, and the actuarial and regulatory questions that surround it are coming into focus at exactly the moment state examiners are building the tools to ask them.

State Farm, USAA, and Allstate together hold approximately 77% of all insurance AI patents, according to research published by Evident AI in late 2025. Within that concentration, Allstate’s portfolio has a distinctive internal logic that this pair makes explicit. The risk-mapping family runs from US11307042, the original three-dimensional risk maps patent, through US11763391, which introduced polynomial multivariable equations to that framework, and now to US12644716, which adds three-dimensional point-cloud representation of discrete risk objects. The ratemaking platform family, now extended by US12632903, shows the analytics-to-production interface being hardened in parallel. From following insurer patent activity over the past eighteen months, that combination, a carrier deepening its risk-representation capability while simultaneously tightening the connection between analytics outputs and rated variables, is an unusual degree of architectural coherence in a field where most patent portfolios read as collections of individual features rather than a designed pipeline.

How the 3D Risk Maps Patent Works

US12644716 operates in three stages. The first is detection: one or more sensors coupled to the vehicle collect information about the surrounding environment. The patent identifies the target objects explicitly, with animals, pedestrians, potholes, and other vehicles among the named categories in the claims. These align with the object classifications that advanced driver-assistance systems routinely produce. The second stage is representation: detected objects are placed into a three-dimensional point cloud, a virtual-world model of the vehicle’s surroundings that captures object location, distance, and spatial orientation rather than reducing the environment to a summary score. The third stage is output: the 3D risk map can be displayed to the driver as a real-time alert system, and it can be retained as a structured record of the driving environment at any point in the trip.

The dual-use architecture is the commercially significant design choice. The patent explicitly provides for the 3D risk map to be leveraged by insurance providers to process insurance claims and for the cost of insurance to be determined based on the risk map. Those are separate functions served by the same artifact. In a claims context, the 3D map from the minutes preceding an accident provides a contemporaneous record of the road environment with better evidentiary properties than a driver’s recollection or a low-resolution dashcam frame: it documents object locations, pavement conditions, and traffic density in spatial terms that support accident reconstruction. In a rating context, the aggregate map built across a policy period, or across a portfolio of policyholders sharing a common route, provides a variable tied to the specific roads an insured actually travels rather than to the rated territory as a geographic average.

The prior generation of Allstate’s risk-mapping patents worked with multivariable equations and aggregate route scores. US11763391, the polynomial risk maps patent, assembled road-surface attributes including slope, wetness, lane count, and construction status; driver profile data; and historical accident records by road segment to produce summary risk values per route. US12644716 operates at a finer level of resolution. Risk objects are represented as discrete, located, classified entities within a spatial model rather than as inputs to a polynomial aggregation. That shift matters for claims reconstruction in particular: a point cloud preserves spatial information that a scalar risk score cannot recover. A pedestrian at a specific location six seconds before impact is a different kind of evidence than a segment-level pedestrian density factor averaged across a route.

Allstate’s telematics program, Drivewise, currently tracks hard-braking frequency, acceleration patterns, speed relative to posted limits, and time-of-day variables, and it offers policyholders discounts up to 40% at renewal based on demonstrated driving behavior. The variables it monitors are conventional telematics dimensions: individual driver actions, not environmental risk objects along the route. The 3D risk maps patent points toward a different variable class, one that characterizes the road environment itself rather than just the driver’s response to it. A driver who regularly passes through a segment with a high density of pedestrian risk objects faces a different exposure distribution than one who does not, independent of how carefully either driver brakes or accelerates. Capturing that environmental component explicitly, rather than leaving it embedded in the geographic territory factor, is what the 3D maps architecture enables.

The Ratemaking Platform Patent: Connecting Analytics to Production

Allstate’s prior ratemaking patents describe the rating infrastructure layer by layer. US9858623 and US9262786 covered composite ratebook configuration, managing the tables, factors, and calculation parameters that feed a rating algorithm across product lines and jurisdictions. US10417711 addressed configuring insurance policy rate routines, the procedural logic that calls those parameters in sequence to produce a premium output. US12632903, the integrated ratemaking platform, sits above that layer: its function is to provide a unified interface through which analytics outputs can flow into the production rating system, reducing the handoffs that have historically separated actuarial model outputs from rated variables in a filed rating plan.

For telematics programs specifically, the integration problem has always been operational as much as technical. The analytics workflow for a usage-based program produces model outputs that must be translated into filed and approved rating factors before they affect a premium. In a typical workflow, that translation involves data exports from the telematics platform, actuarial analysis of the outputs, preparation of rate revision filings for each state, state regulatory review and approval, and system update to push the revised factors into the production rating engine. An integrated ratemaking platform compresses several of those intermediate steps by providing a structured path through which model outputs can move toward rated variable status more directly. It does not eliminate the actuarial and regulatory requirements; those remain necessary before any variable affects a premium. It eliminates the scaffolding that has made the journey slow.

The practical consequence for the 3D risk maps architecture is this: if sensor-derived, object-level route risk scores are to become rated variables in a personal-auto program, the integrated ratemaking platform provides the production-side infrastructure to receive them. The risk maps patent generates the evidence artifact. The ratemaking platform patent provides the integration point. Both pieces are now owned and granted, and both carry the legal documentation of conception and reduction to practice that patent grants represent.

The Actuarial Support Questions Route-Level Scoring Raises

Any rating variable that affects premium must survive actuarial support review when filed with a state regulator. For conventional telematics variables, carriers have accumulated several years of correlated loss experience. For route-level and object-level risk scores derived from 3D spatial maps, the actuarial support problem has more layers than anything the prior generation of telematics filings required.

ASOP 56, Modeling, effective October 2020, requires actuaries responsible for a predictive model to document the model’s design and limitations, characterize sensitivity to key assumptions, and disclose material limitations inherent in its application. A 3D risk map model involves at minimum four distinct parameter layers: sensor calibration thresholds that determine when a signal qualifies as a classified object; object classification algorithms that distinguish a pedestrian from background environmental noise; risk value assignment functions that translate an object’s presence, distance, and relative velocity into a risk contribution; and route-aggregation methodology that combines object-level contributions into a policy-period score usable as a rating variable. Each layer carries its own distributional assumptions. ASOP 56 requires those assumptions to be characterized and their material sensitivity analyzed, not simply held constant and undisclosed.

ASOP 25, Credibility Procedures, adds a further dimension. Route-level risk scores in jurisdictions or risk segments with thin data carry less statistical credibility than scores built from high-volume routes. An insured who travels a given road segment twice a week generates less segment-level experience than one who drives it twice daily. How credibility weighting flows through a 3D map aggregation is a question actuarial support documentation will need to answer explicitly. Blend the observed route risk score with a complement-of-credibility based on what class? The rated territory? A statewide road-type average? The complement selection is an actuarial judgment call with material premium implications, and it belongs in the actuarial memorandum supporting any filing that uses this variable.

The deeper question sits underneath both the sensitivity and credibility layers: whether a route-level environmental risk score is a direct predictor of future losses for the individual insured, or whether it is partly a proxy for something else. A driver who regularly passes through construction zones or high-pedestrian-density corridors does face elevated segment-level exposure during those trips. But the correlation also runs through geography, economics, and schedule. Drivers whose primary routes pass through deteriorated road surfaces or high-density urban environments may be concentrated in lower-income neighborhoods, may work jobs with irregular hours that require late-night travel on high-risk roads, or may live in areas with historical loss patterns driven by infrastructure quality rather than individual driving behavior. Actuaries working with the American Academy of Actuaries framework on algorithmic accountability have developed guidance on identifying when a variable that appears predictive is transmitting a socioeconomic signal rather than capturing individual risk directly. That framework is the professional reference point for any actuary assessing whether a sensor-derived route variable adds predictive power that belongs in a rating plan or introduces a proxy discrimination concern that does not.

These are solvable problems. Carriers filing telematics variables have developed actuarial support methodologies for objection-vulnerable predictors before, and rating plans can include algorithmic guards that prevent the worst proxy effects. But the solutions require documentation, testing, and disclosure that takes time to build, and that work must happen before a variable derived from the 3D maps architecture can be positioned in a state filing as actuarially supported.

What State Rate Filing Reviewers Will Ask

When a carrier introduces a new rating variable derived from sensor data, the filing becomes a representation to the regulator that the variable is actuarially justified, does not produce prohibited discrimination, and is explainable in sufficient detail to permit regulatory review. For sensor-derived object-classification scoring, all three of those representations require more documentation than conventional telematics variables have needed, and the regulatory environment is becoming more demanding on precisely this category of evidence.

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in December 2023 and distributed to state departments for adoption, asks insurers to document AI system governance, risk management, and use protocols, and to make that documentation available to examiners on request. For rate filings, states that have adopted or reference the bulletin’s standards will expect supporting material beyond historical loss ratios. Examiners may ask for the sensor calibration basis, a description of the training data used to develop the object classification algorithm, validation results from a hold-out test set, and a bias analysis confirming the variable does not produce prohibited discrimination across protected or legally sensitive classes. That is a different evidentiary package than a five-year loss development triangle correlated against a binary telematics segment indicator.

Twelve states are running NAIC AI evaluation pilot market conduct examinations through September 2026. The pilot is building a standardized examination template for reviewing AI governance programs during market conduct exams, with structured exhibits covering model inventory, documentation quality, third-party vendor oversight, and bias testing protocols. The findings from that pilot will shape what every other state department expects when it sees a sensor-derived variable in a rate filing. Actuaries supporting personal-auto programs that deploy object-classification risk scoring should read the pilot’s published guidance, particularly the exhibits addressing model documentation and data quality governance, as the floor of what state regulators will require rather than a ceiling.

One structural question the patent pair introduces for state rate filers is whether the integrated ratemaking platform itself falls within the scope of what state departments are examining under the AI Model Bulletin framework. If the platform functions as an automated system that produces rating outputs downstream of model scores, it may qualify as an AI system subject to governance and disclosure requirements that go beyond traditional actuarial certification. The distinction between a rating engine that applies filed factors and an AI system that generates rating variables is becoming a live regulatory question, and the integrated ratemaking platform patent sits in the space between those definitions.

Privacy, Explainability, and Where the Controls Must Sit

The 3D risk maps patent involves continuous or trip-level collection of detailed spatial sensor data about the vehicle’s surroundings. The data volume and specificity involved creates privacy obligations that go beyond what traditional telematics programs generate. The California Consumer Privacy Act and its amendments, along with analogous statutes in a growing number of states, impose consent, disclosure, and data-retention constraints on personal information derived from location and sensor data. Several states have enacted or proposed separate laws governing biometric and spatial data that may apply to point-cloud representations of a vehicle’s environment, particularly if those representations incidentally capture identifiable individuals in the sensor field of view. The patent’s explicit claim of using the 3D map for claims processing and premium determination makes the data collection difficult to classify as incidental telemetry. It is a stated input to insurance decision-making, which triggers the heightened regulatory scrutiny that applies to adverse underwriting information under state fair information practice statutes.

Explainability requirements add a separate compliance layer. The NAIC Model Bulletin’s transparency principle asks that AI system decision-making be understandable to regulators and consumers. A three-dimensional point-cloud risk score is not inherently explainable in those terms. When a policyholder’s premium is affected because the system detected a sustained concentration of pedestrian risk objects along their primary commute corridor, the carrier needs a disclosure framework that can communicate that basis accurately, in language a consumer complaint review or a state regulatory examination can evaluate. Translating a spatial risk model output into a premium-explanation statement is a different engineering and legal problem than explaining a hard-braking frequency discount, and it requires a document trail that begins in the patent’s architecture and runs through actuarial support, regulatory approval, and policyholder communication.

The controls that neither patent prescribes, explainability frameworks, bias testing protocols, consent documentation architecture, and data-retention governance, are exactly what the actuarial and regulatory infrastructure requires the carrier to add before the patented architecture can function as a legally compliant, filed, and approved rating system. The patents describe what the mechanism can do. The actuarial professional standards and the regulatory framework determine what must surround it before it can be used to set premiums.

Three Markers Worth Watching

Personal-auto rate filings in states where Allstate has expanded Drivewise will be the first public record of how sensor-derived risk scores are being positioned actuarially. Any filing introducing a new telematics variable category after mid-2025, particularly one framed in terms of route-level environmental characteristics rather than individual driver actions, should be read against US12644716’s claims to assess whether the variable is downstream of the 3D maps system. Rate filing databases maintained by state departments and SERFF filings where available are the primary public record; Massachusetts, California, and New York require detailed actuarial memoranda that will be the most informative documents when they appear.

The NAIC AI pilot examination findings, expected to be summarized in fall 2026, will establish the first structured template for what state regulators want from insurers using sensor-derived rating variables during market conduct exams. If the pilot produces a published report of examination observations, it will be the most significant regulatory guidance document for this technology category since the December 2023 Model Bulletin itself. The specific documentation requests that pilot examiners have found inadequate in practice, and the model inventory and validation standards that have satisfied them, will define the regulatory bar.

The actuarial profession’s developing guidance on bias in pricing and marketing systems will set the professional standard that actuaries must apply when validating a route-based risk variable for rate filing purposes. The Academy’s Property and Casualty Committee on Equity and Fairness has been building out issue briefs specifically on bias in underwriting and claims administration. When that guidance reaches recommendation status, it will define the professional obligation for any actuary supporting a filing that uses route-level sensor scoring as a rated variable. Filings made before that guidance exists face a different evidentiary standard than those made after it is formally adopted.

The patent grants are real, and the architecture they describe is coherent. Allstate now holds both the evidence-generation system and the analytics-to-production integration point as documented intellectual property, filed and prosecuted through the USPTO. The question that will determine whether sensor-derived 3D risk maps become a mainstream personal-auto rating variable is not whether the technology functions; it clearly does. The question is whether the surrounding actuarial validation, regulatory documentation, and privacy compliance infrastructure can be built to the standard these variables will require. Those are difficult but tractable problems. Carriers that solve them first will have both the IP and the regulatory pathway. Carriers that follow will have neither exclusive to themselves.