A Claims AI Patent Now Reaches Into Loss Reserving

US Patent 12,694,454 B2, granted to Assured Insurance Technologies on July 28, 2026, decides whether a wrecked vehicle is repairable or totaled and routes the file accordingly. It also names a "reserve estimation module" that turns the same inputs into the dollar figure the carrier holds against the claim.

That second function moves a claims-workflow tool onto the loss reserve schedule. Assured is not a carrier.

Key Takeaways

US 12,694,454 B2 issued July 28, 2026 on an application filed April 24, 2025, itself a continuation of an August 19, 2024 filing. Claim 1 outputs repairable or totaled; claim 6 adds salvageable as a third state.
A named "reserve estimation module 170" generates the total cost of a claim from police reports, medical reports, fraud-detection results and collision simulations, and adjusts as the claim file builds rather than fixing a number at intake.
The training loop is a reserve-adequacy backtest. The model reads closed files carrying both the initial reserve estimate and the final payout, then tunes its parameters to shrink the discrepancy between the two.
23.1% of all 2025 US auto claims were total losses on CCC Intelligent Solutions' count, a record, which sizes the volume of claims where a misclassification now lands on a case reserve.
Two Assured filings claim the same module, one granted and one published as US 2026/0050991 A1, so the reserve estimate is being patented as an invention rather than as a byproduct of triage.

What US 12,694,454 B2 Claims

The patent, titled Machine-Learning Method of Loss Prediction Using Incident Information, issued July 28, 2026 from an application filed April 24, 2025, itself a continuation of an application filed August 19, 2024. Its abstract states the function plainly: incident data tied to a vehicle collision goes in, and a total loss prediction "indicative of whether the vehicle is repairable or totaled" comes out.

Claim 1 describes more than a binary call. The system trains on historical vehicle incident data using discrepancy analysis, presents a three-dimensional representation of the vehicle so a user can mark damage locations, and predicts whether the vehicle is repairable or totaled relative to a value. A human still marks the damage on that model, so the system does not work from raw photographs alone. Claim 6 widens the output to three states by adding salvageable, for a vehicle worth more stripped for parts than sold intact to a body shop.

What the model decides is what the file does next. Each output carries its own claimed downstream action, which is where a classification error stops being a modelling question and becomes an operational one.

Model output	Downstream action claimed
Totaled	Coordinate a tow service to a scrap yard
Repairable	Select a service provider; schedule repair
Salvageable (claim 6)	Distinct third state alongside repairable and totaled

The Reserve Estimation Module and the Case Reserve It Sets

The classifier's output does not stay inside the workflow layer. The specification names a "reserve estimation module 170" that ingests the full corpus a claim accumulates: contextual claim data, damage specifics, injury reports, medical reports, police reports, fraud-detection results, vehicle-collision simulations and guided content capture evidence. From that corpus it generates the total amount of cost arising from the claim event, covering medical, repair, replacement and service-provider costs.

The training method is the part a reserving actuary should read twice. The model processes closed claim files that carry both an initial reserve estimate and the final payout, identifies the discrepancy between them, and fine-tunes its parameters to shrink that gap on the next prediction. That is a reserve-adequacy backtest running inside a training loop rather than inside a development triangle. The estimate also adjusts as a claim's corpus builds, which makes it a running case reserve rather than a number fixed at intake. A second filing, US 2026/0050991 A1, published February 19, 2026, claims the same architecture standing alone.

The dollars follow the classification, not only the severity estimate. A total-loss claim converges to actual cash value within weeks once ownership of the wreck transfers. A repairable claim carries supplemental damage found when a panel comes off, rental extension on back-ordered parts, and sometimes a mid-repair reclassification once supplements cross a state threshold. Call a claim repairable that should have been totaled and the case reserve is not merely light: a $4,000 repair that later re-flags as a $14,000 total loss inherits a settlement timeline, salvage recovery entry and closure pattern the original reserve never modelled.

With CCC putting 2025's total-loss share at a record 23.1% of US auto claims and driveable-vehicle total losses more than doubling since 2021, that is a large and growing share of any personal auto book. A systematic miscall in one direction does not net out in the aggregate. It shows up in the paid-to-case diagonal, which is the signal reserving actuaries select IBNER factors from, as an earlier look at claims-handling speed in workers compensation found in a different line.

A Case Reserve Set Inside Code the Carrier Does Not Own

The governance question is not whether the model is good. It is where the number enters the carrier's books, and who can see how it was made.

Assured is a vendor. The Palo Alto company, founded in 2019 by Theo Patt and Justin Lewis-Weber, builds first-notice-of-loss, messaging, fraud-detection and catastrophe-response tools it says process tens of millions of claims a year for top insurers (Assured), and it reached a $1 billion valuation on April 3, 2025, with 199 employees as of June 30, 2026 (Tracxn). Its patent fences off an architecture it can embed across every carrier that buys the platform, unlike the USAA and Allstate grants built on a single carrier's own data and deployed inside its own operations.

A rating factor faces a filing requirement because a regulator needs the mechanics before the number reaches a policyholder's bill. A case reserve carries no equivalent step. It lands on the loss and loss-adjustment-expense reserve schedule and is reviewed after the fact, by the actuary signing the statement of opinion and by auditors. ASOP No. 43 directs that actuary to evaluate the data underlying an unpaid claim estimate and to account for changes in the claims process that could affect its reliability.

That standard was adopted in June 2007, nearly two decades before a machine-generated case reserve existed, but swapping adjuster judgment for a vendor model at first notice of loss is exactly the claims-process change its data provisions were written to catch. Three questions follow, and the public claims answer none of them: whose closed files trained the discrepancy loop, one carrier's book or a pool spanning Assured's roster; how often the model retrains; and whether a carrier sees a retraining before it changes the reserves flowing onto its books.

None of that requires assuming anything about the vendor, and none of it is unique to Assured. It is the audit trail a case reserve generated inside licensed code requires, and that one set by an adjuster's judgment never did.

A Claims AI Patent Now Reaches Into Loss Reserving

Key Takeaways

What US 12,694,454 B2 Claims

The Reserve Estimation Module and the Case Reserve It Sets

A Case Reserve Set Inside Code the Carrier Does Not Own

Further Reading

Sources