A UnitedHealth Patent Predicts the Claim, Then Sells the Coverage

A machine learning model predicts the probability that a health plan member will need a particular medical service. The system then shows that member an add-in coverage option for exactly that service, takes the election, and stores it. That sequence is claim 1 of US Patent 12,711,553 B1, granted to Bind Benefits, Inc. on August 18, 2026 (USPTO Official Gazette, Vol. 1549).

Bind Benefits does business as Surest, the copay-only, deductible-free plan UnitedHealthcare sells to employers. The carrier's own disclaimer states that for self-funded plans, administrative services "are provided by Bind Benefits, Inc. d/b/a Surest" (UnitedHealthcare). A sibling patent, US 12,705,672 B1, issued a week earlier on August 11. A third application in the family reached Patented Case status on August 26, which queues a fourth grant.

Key Takeaways

• 20 claims in US 12,711,553 and 17 in US 12,705,672, both classified G06Q 40/08 for insurance data processing and both prosecuted in art unit 3695 by the same examiner, Ayal I. Sharon (USPTO Open Data Portal, August 2026).
• Nine years and two months separate the June 23, 2017 filing of application 15/632,052 from its August 18, 2026 grant, while its own 2019 continuation-in-part granted a week earlier. Prosecution in class 705/4 does not run in filing order.
• A coverage amount field and a pointer to "callable options" sit inside claim 1's Personal Protection Plan data structure. The claims define the option and the election. No limitation reprices the add-in from the probability the model has just computed.
• 21.5 million fee-based commercial members at December 31, 2025, up from 20.9 million a year earlier (UnitedHealth Group, January 2026). That self-funded book is where an elected add-in's claims land on the employer rather than the carrier.
• 88.9% adjusted medical care ratio for full-year 2025, 340 basis points above 2024 (UnitedHealth Group, January 2026). Administrative-services business does not pass through that ratio at all.

Patent Details

Field	US 12,711,553 B1	US 12,705,672 B1
Title	Use Determination Risk Coverage Datastructure for On-Demand and Increased Efficiency Coverage Detection and Rebalancing Apparatuses, Methods and Systems
Application	15/632,052	16/659,438
Filed	June 23, 2017	October 21, 2019
Granted	August 18, 2026 (OG Vol. 1549 No. 3)	August 11, 2026 (OG Vol. 1549 No. 2)
Pendency	9 years, 2 months	6 years, 10 months
Assignee	Bind Benefits, Inc.	Bind Benefits, Inc.
First inventor	Anthony Miller	Anthony Miller
Claims	20	17
Classification	CPC G06Q 40/08, G06Q 10/067; USPC 705/4
Examiner / art unit	Ayal I. Sharon / 3695
Earlier family members	US 11,790,454 (granted 2023); US 12,266,018 (granted 2025)

What Claim 1 Covers

The Surest design splits a plan into a core coverage component and what the patents call atomized add-ins: coverage for "a discrete provider, discrete medical service, or discrete medical condition that is distinct from the healthcare core coverage component" (US 12,711,553, claim 1). The member holds the core through the year and attaches the pieces.

Claim 1 of the August 18 grant runs as an ordered sequence of operations:

• Receive an event signal data structure indicating an event involving an enrolled member, plus a member data structure carrying at least an identifier.
• Predict, by one or more machine learning processes, "a probability that a medical condition or service will occur in association with the healthcare plan member" from that event signal.
• Generate, from that probability and the set of add-ins available for addition to the plan, a recommendation containing a first add-in for that member.
• Present it through the enrollment user interface as a Personal Protection Plan data structure with a coverage amount field and a pointer to callable options.
• Receive the member's selection and store it.

The earlier grant covers the engine underneath. US 12,705,672 runs 17 claims that locate a member's position on a treatment path, weigh how likely a given provider is to use each available path, and select a provider on longitudinal treatment value and copay calculations. It is a continuation-in-part of the 2017 application, and it issued first.

Five applications sit in this family and four have now issued, all to the same examiner. The oldest took more than nine years to clear an art unit that spent the post-Alice decade rejecting insurance business methods as abstract ideas. The surviving claim recites machine learning processes, a named data structure, and a user interface, which is the shape a 705/4 claim now has to take.

The Mechanism: An Election Triggered by a Prediction

The Society of Actuaries defines adverse selection as "the greater tendency of people with a more than average likelihood of loss to apply for or continue insurance, when compared with other people" (SOA Actuarial Glossary, 2022). The textbook version is an asymmetry the insurer cannot see. Here it is computed by the plan's own model and handed to the member with a purchase button attached.

That relocates the pricing problem rather than removing it. Flexible benefit pricing sets a loading factor for each design and derives subsidization factors across the options, so the design attracting healthier lives helps fund the one attracting sicker lives. Those loads are calibrated on an annual election with a twelve-month lock, and the lock is what gives the load an exposure period to amortize across.

An event-triggered add-in has no such period. The member elects in the window between the signal and the service, so the conditional expected cost of the elected add-in is the model's predicted probability multiplied by the cost of the service, not the pooled unconditional rate across the enrolled group. Charge the pooled rate into that election and the add-in's loss ratio at the point of sale converges on one by construction.

The claims do not close the gap. The specification calculates an "insurance cost ... associated with the selected atomized procedure for the selected provider or provider network" and a pay period deduction against an available HRA balance. That is a cost display keyed to the procedure, not a premium keyed to the prediction that produced the offer. The one lever the patents do supply points the other way: steering members onto lower-cost treatment paths is the economic offset, and UnitedHealthcare markets the design on up to 11% lower employer costs, its own figure.

The ACA individual market fences this behavior deliberately. Special enrollment periods are gated on qualifying life events precisely to stop mid-year election on new health information, and carriers still load for the residual, as this year's 2027 morbidity work shows. The employer group setting has no equivalent guard.

Where the Selection Cost Lands

Surest's nationwide footprint is self-funded, with fully insured and level-funded variants underwritten through UnitedHealthcare Insurance Company and All Savers. In the self-funded majority the carrier collects administrative fees and the plan sponsor's fund absorbs the claims. An add-in elected two weeks before a predicted procedure is paid out of the employer's account.

That routing is why the design's morbidity does not surface in the numbers most readers watch. UnitedHealth Group's adjusted 2025 medical care ratio was 88.9%, up 340 basis points year over year, and its Employer and Individual segment booked $79.2 billion of revenue while people served fell by 80,000, with "growth in employer self-funded offerings more than offset by attrition in both group fully-insured and individual products" (UnitedHealth Group Form 8-K, January 2026). The growing book is the one whose claims never enter the ratio.

The harder problem is the reserve. Incurred-but-not-reported estimates run off completion factors fitted to historical payment lag, and the fit assumes enrolled exposure is independent of incurral timing. An add-in triggered by a predicted-condition signal breaks that assumption on purpose. Members attach coverage in a given month because a model said the service was likely to follow, so enrollment and incurral become correlated and a triangle built on aggregate member months understates unpaid liability at the short durations where these elections concentrate.

Stop-loss carries the same defect a layer up. Attachment points are fixed before the plan year against a defined benefit schedule and an expected claims distribution, and a benefit the member can add mid-year moves the covered-services definition after the attachment point is set. The employer bought this plan for its price transparency and now holds the cost basis of an option it did not write, funded by a reserve calibrated on a population that could not buy coverage when it expected to use it.