Ford Patents a UBI Rating Engine for a Data Pipeline It Shut Off

Ford Global Technologies won US Patent 12,694,456 on July 28, 2026, for a system that runs high-dimensional Bayesian optimization against an archive of vehicle signals and their insurance effects, builds weighted aggregation functions from the result, and pushes those functions into vehicles so they compute a composite usage-based insurance score on board rather than transmitting raw driving data.

Ford filed the application on February 12, 2024, the same year it wound down the consumer telematics-sharing program the method was designed to run.

Key Takeaways

• Claim 5 sets weights below a threshold to zero, dropping low-relevance signals from the aggregate entirely, while claim 4 assigns each remaining signal a weight for its importance to the predicted outcome.
• The vehicle computes the composite and transmits only the aggregate. The raw signal stream never reaches an insurer's ingestion pipeline, and whoever controls the firmware controls what leaves the car.
• $12.75 million was California's settlement with General Motors over selling driving and location data to brokers between 2020 and 2024, brought on data-minimization and purpose-limitation grounds.
• Texas requires a carrier to disclose the vendor, version, inputs and use of any third-party model behind a filed rate. Claim 1's loop can reoptimize the weighting between filing cycles.

What Claim 1 Actually Builds

The independent claims describe a two-stage system. A server-side process performs multi-dimensional Bayesian optimization over a data archive containing telematics signals and their observed effect on insurance outcomes, iterating "until a convergence criterion is met and/or until an iteration budget is exhausted." Each iteration trains a surrogate model, evaluates an acquisition function, and tests a new weighted grouping of signals against the UBI rating model, generating a fresh input-output pair for the next round.

Claim 4 specifies that each resulting aggregation function "applies respective weights to each of the plurality of signals," the weight standing for that signal's relative importance to the outcome being predicted. Claim 5 adds that weights falling below a threshold are set to zero.

Actuaries will recognize the shape even if the vocabulary is unfamiliar. Selecting which telematics variables belong in a rating algorithm, and how much weight each deserves, is a factor-selection exercise every UBI pricing team already runs by hand or through a generalized linear model, on a cadence tied to a filed rate revision. The claims automate that selection as a closed loop that reoptimizes without a scheduled model build.

The second stage carries the sharper competitive edge. The vehicle, not the insurer's ingestion pipeline, computes the weighted composite locally and transmits that aggregate rather than the underlying signal stream, a design the patent frames as reducing vehicle-to-cloud data transfer.

Data Minimization Becomes a Rating Architecture

Before late 2024, a Ford owner who wanted usage-based insurance opted in twice, once with an insurer and again on the vehicle's own screen, before Ford would relay telematics to that carrier. Ford discontinued the arrangement in 2024 (Ford Authority, March 2024), after reporting that General Motors had been piping OnStar Smart Driver location and driving-behavior data to brokers who resold it to insurers, in some cases collecting geolocation as often as every three seconds.

That resolved into hard regulatory outcomes. The FTC finalized its consent order against GM and OnStar on January 14, 2026 by a 2-0 commission vote, barring sales of geolocation or driver-behavior data to any consumer reporting agency for five years and requiring affirmative, express consent for the full twenty-year life of the order (Federal Trade Commission). California followed with a $12.75 million settlement over sales to Verisk Analytics and LexisNexis Risk Solutions between 2020 and 2024, business worth roughly $20 million to GM nationwide (California Department of Justice).

Ford's architecture is built around the principle those cases turned on. A vehicle that computes a weighted composite on board and transmits only the aggregate structurally cannot be the subject of a complaint that it sold second-by-second location pings, because the raw trace never leaves the car. Whether that was designed for legal cover or for bandwidth cost, the effect is the same.

The pricing consequence is where the aggregation layer ends up sitting. GM's model was automaker as raw-data supplier to the two firms that dominate third-party telematics scoring (NAIC Center for Insurance Policy and Research). Ford's grant supports owning the rating output instead. Ford Pro Insure, its commercial-fleet product built with Pie Insurance, runs through The American Road Insurance Company, live in Arizona, Illinois, Indiana, Tennessee and Wisconsin. A carrier writing usage-based policies on Ford vehicles may find the weighting step, which an actuary would ordinarily control end to end, inside Ford's claims rather than its own model.

The grant is also narrower than the broad scoring application Ford was reportedly pursuing in 2022 (Repairer Driven News), which does not appear among its granted patents, a pattern consistent with the sharper Section 101 eligibility bar applied to software claims this year. It sits in a different layer again from LexisNexis Risk Solutions' six telematics patents through 2018, covering ingestion, normalization, storage and distribution across what it then put at 60% of global vehicle production (LexisNexis Risk Solutions).

Patent or Portfolio	Holder	Layer of the Stack	What It Actually Covers
US 12,694,456 (2026)	Ford Global Technologies	Rating math	Bayesian-optimized weighted aggregation of telematics signals into a UBI rate-prediction input, computed on the vehicle
US 11,257,146 (2022)	State Farm Mutual	Product/use case	Incentives and penalties applied to vehicle renters based on telematics data during a rental period
Six-patent portfolio (through 2018)	LexisNexis Risk Solutions	Data pipe	Ingestion, normalization, storage, and distribution of connected-car data across manufacturers
2022 application (unresolved)	Ford Global Technologies	Broad scoring concept	Driver/vehicle/context risk score from ADAS activation data for third-party aggregators; does not appear among Ford's granted patents

Whose Documentation Backs the Rate

Rate-filing regulation was not built for a pricing input a carrier licenses from a patent it does not own and cannot see inside. Texas already requires carriers to disclose the vendor's name, the model's name and version, a description of the model, its inputs, and how the output is used in ratemaking or underwriting, along with the rating variables that depend on that output (Texas Department of Insurance). That regime was designed around static, vendor-supplied catastrophe and predictive models.

Claim 1's aggregation function is not static. The Bayesian loop can reoptimize the signal weighting against Ford's own archive between filing cycles, and neither a carrier licensing the output nor a state examiner reviewing the filing has visibility into the archive, the acquisition function or the convergence criterion, because those details sit inside a method claim rather than a disclosure document an examiner can audit line by line.

The NAIC's telematics and big data work already flags the transparency gap that arises when a UBI score comes from a vendor a carrier does not fully control (NAIC Insurance Topics). This sharpens it rather than resolving it, because the vendor here is also the manufacturer with first-party sensor access no app or plug-in dongle competitor can replicate, and because continuous automated reoptimization is a design feature of the claim rather than an edge case.

A carrier licensing the function therefore has to reconstruct, for its actuarial memorandum, a description of a scoring methodology that changes on Ford's schedule rather than its own. That is a different documentation problem from naming a vendor model version, and nothing in the current filing regime anticipates it.