A Quanata Patent Turns Driving Data Into a Credit Score

Quanata, LLC, the State Farm-backed telematics developer formerly called BlueOwl, has a granted US patent on a model that reads accelerometer, GPS, and gyroscope data from a driver's trips and returns a predicted credit score for that driver. US Patent 12,711,543 B2 issued on August 18, 2026 (USPTO Official Gazette, Vol. 1549). Claim 2 lets the resulting score be sent to a bank, a credit card company, or a prospective employer.

The application was filed on December 17, 2023, two days before its parent issued, and it took two rejections, a request for continued examination, and a terminal disclaimer to get through. The examiner recorded that the new claim 1 "is broader than the claim of the patent" it descends from. This was a deliberate widening, not a housekeeping refile.

Key Takeaways

• Twenty claims, three of them independent, covering a device, a method, and a computer-readable medium. Claim 6 permits the predicted figure to overwrite the user's own credit score once a modeled error probability drops below a threshold.
• A single limitation carried the patent over the prior art: that the probability of error falls as the number of telematics records rises. The examiner found every other element of claim 1 already taught by three earlier publications.
• A November 10, 2025 final rejection held the whole claim set ineligible under Section 101 as a method of organizing human activity. Allowance followed on April 14, 2026, after the applicant wrote the physical sensors into the claim body.
• Primary classification is G06Q 40/03, the CPC bucket for credit and loans, ahead of G06Q 40/08 for insurance. The Patent Office reads this as a credit invention that happens to run on driving data.
• 29.7 million US adults hold unscored credit records and another 13.5 million are credit invisible (CFPB, June 2025), which is the population a driving-derived score would be sold into.

Patent Details

Patent number	US 12,711,543 B2
Title	Systems and Methods for Generating a Credit Score Based at Least in Part Upon Telematics Data
Applicant	Quanata, LLC (San Francisco, CA)
Inventor	Michael Sungjun Kim
Application	18/542,716, filed December 17, 2023
Granted	August 18, 2026 (Official Gazette Vol. 1549)
Parent	US 11,847,691 B2 (app. 16/803,318, filed February 27, 2020, granted December 19, 2023, applicant BlueOwl, LLC)
Examiner / art unit	Edward J. Baird, Art Unit 3692
Claims	20 (three independent)
Classification	G06Q 40/03 (credit, loans), G06Q 40/08 (insurance), G06N 20/00 (machine learning), G07C 5/008 (vehicle data recording)

What Claim 1 Covers, and What It Took to Get It

Claim 1 collects telematics from one or more accelerometers, global positioning sensors, or gyroscopes during operation of the vehicle, then applies a trained model to that stream plus "current user data" to predict a credit score. The training set is the interesting part: historical credit scores, user data, and telematics drawn from users who share similar driving characteristics and whose credit scores fall within a predetermined range.

That construction is a matched-cohort design. The model is not learning a general relationship between driving and creditworthiness; it is learning what a credit band looks like when its members drive. Claim 4 lists what else rides along in the training data: income, education level, age, gender, and occupation.

Getting here was not smooth. The examiner's final rejection of November 10, 2025 held claims 1 through 20 directed to the abstract idea of "determining a probability of error associated with the credit score," grouped under methods of organizing human activity, with the sensor language dismissed as "merely descriptions of data" that "do not impose any meaningful limits on the computer implementation of the abstract idea."

Then the examiner supplied the fix in the same document, suggesting the applicant add "GPS device, an accelerometer, a gyroscope" as recited components rather than data labels. Quanata amended accordingly, filed a terminal disclaimer on March 24, 2026 to clear the double-patenting rejection over the parent, and had a notice of allowance three weeks later. The eligibility of a credit-scoring model turned on naming the hardware.

The Credibility Limitation

The reasons for allowance are unusually precise about what is new. Three prior publications together taught the sensors, the model, the matched training cohort, and the retraining loop. What they failed to teach was one clause: determine a probability of error "based upon a number of records in the current vehicle telematics data, wherein the probability of error is lower when the number of records is higher."

That is a credibility statement written in patent grammar. It is the same idea an actuary applies when the standard error of an estimate shrinks with exposure volume, and it is the reason a classical credibility weight rises toward one as claim counts accumulate. Quanata's patentable core is not the driving-to-credit mapping at all. It is the rule that says how many trips buy how much confidence in the answer.

Claim 6 turns that into a threshold. Once the modeled error probability falls below a set level, the system updates the user's credit score using the prediction. In actuarial terms this is a full-credibility standard, expressed as a volume trigger for when a modeled figure stops being an indication and becomes the figure of record.

For a rating plan the consequence is correlation. A personal auto tier that already carries a telematics score and a credit-based insurance score would now hold a third input built from the same sensor stream as the first and calibrated against the second. Fitting all three in one rating model without collapsing the credit signal into a driving signal is a multicollinearity problem before it is a fairness one.

It also lands in books that already have the trip volume to clear the threshold, as Progressive's 21 million telematics policyholders indicate. Quanata's contents-valuation grant two weeks earlier came from the same drafting shop, and the pattern in both is a model that infers a number the insurer used to buy from a vendor.

A Score No Bureau Produced

The complication is that the output has no furnisher behind it. A credit-based insurance score comes from a consumer reporting agency, and the machinery around it assumes that: dispute rights, file access, and the adverse-action notice required by 15 U.S.C. 1681m when a score contributes to a denial or a worse rate. A number inferred from cornering forces has no tradeline to correct and no bureau file to inspect.

Claim 2 makes this sharper by naming a prospective employer as a recipient. Employment screening and credit extension are the classic permissible purposes under the Fair Credit Reporting Act, and a party assembling consumer data into a score furnished to third parties for those decisions is describing a consumer reporting agency subject to Regulation V. The patent claims the capability; it does not claim compliance with it.

State law then splits the input from the output. The NAIC notes that "some groups allege that the use of credit-based insurance scores falls disproportionately on certain minority and low-income groups," and California, Hawaii, Massachusetts, and Michigan bar the scores in personal auto outright, with a congressional probe pressing the question further. Yet California Insurance Code 1861.02(a) makes annual miles driven the second mandatory rating factor, behind driving safety record. The GPS data the patent ingests is compulsory in California rating; the credit score it produces is prohibited.

Disparate impact is where the two halves meet. The CFPB's corrected estimate puts 13.5 million adults at credit invisible and 29.7 million more with unscored records, revised in June 2025 from 25.9 million and 17.2 million after the Bureau found deferred student loans and collections-only files missing from its panel. Credit invisibility is not evenly distributed, and a model trained on gender, income, and education to reproduce the credit bands of drivers who already have scores will carry whatever gradient those bands contain. The dispute rights that exist to catch it attach to the bureau file, and this score does not have one.