WISeR, Explained

Traditional Medicare has historically used prior authorization only in narrow, service-specific programs — repetitive non-emergent ambulance transport, certain outpatient department services. On July 1, 2025, CMS published a Federal Register notice announcing something broader: the Wasteful and Inappropriate Service Reduction (WISeR) Model, a six-year Innovation Center test that brings “technology-enabled prior authorization” — CMS names artificial intelligence and machine learning explicitly — into Medicare fee-for-service for a defined list of services, in six states, starting January 1, 2026.¹ The reviewers are not the Medicare Administrative Contractors. They are private companies, selected for their experience running technology-assisted prior authorization for other payers, and they are paid from the savings their reviews produce.¹

What follows is the model as the documents describe it — the mechanics, the service list as amended, the payment design, and the questions the published record leaves open.

Prior authorization under WISeR is technically voluntary, but declining it has a consequence. The notice lays out three paths.¹ A provider can submit a prior authorization request directly to the model participant, or to their MAC, which routes it to the model participant. If the request meets the coverage, coding, and documentation requirements already in the governing statute, rules, NCDs, and LCDs, the participant issues a provisional affirmation and coordinates a unique tracking number; a correctly billed claim carrying that number is paid in full. If the provider skips prior authorization entirely, the claim is flagged for pre-payment medical review by the same model participant, with documentation requested after the fact.

A non-affirmation must come with a detailed reason. The provider can resubmit an unlimited number of times, and on resubmission can request a peer-to-peer review to inform the new determination.¹Expedited review exists where the standard timeframe could jeopardize life or health, though CMS expects such requests to be “extremely rare” given the nonemergent target list. And critically for anyone who works denials: a non-affirmed request does not bar the claim. Submitting it anyway produces a denial that is an initial determination under 42 CFR part 405, subpart I — the standard appeal ladder applies, and the notice states plainly that “the existing claims appeals process would not change under the model.”¹

The documentation burden is not new, only earlier. CMS's own framing: the model “will require the same information and clinical documentation that is already required to support Medicare FFS payment but earlier in the process, namely, prior to the service being furnished.”¹There is also an exit ramp — a “gold carding” exemption CMS is exploring for providers who sustain a provisional-affirmation rate of 90 percent or better during periodic assessment.¹

The July 2025 notice named the services by their coverage authority — the NCD or LCD family that already defines when Medicare pays.¹Two of them never went live: an April 6, 2026 notice delayed deep brain stimulation and percutaneous image-guided lumbar decompression “to allow additional time for operational readiness,” to a future date to be announced.² A companion correction in the same issue fixed technical errors in the original LCD lists,³ which is why the table below counts the LCD families rather than reproducing every L-number — the authoritative list is the docket itself.

The model participants are companies “with experience implementing technology-enhanced prior authorization with other payers, including MA plans.”¹ The technology screens; it is not, on paper, permitted to say no by itself. CMS's fact sheet draws the line in one sentence:

Because these companies handle protected health information on behalf of the Medicare program, the notice makes them HIPAA business associates of CMS, bound by business associate agreements.¹

This is the model's most consequential choice, and CMS describes it without euphemism: participants “would be paid under a novel payment approach where the model participants are compensated based on a share of averted expenditures” — “a share of averted expenses in lieu of the traditional acquisition-based approach.”¹ In plainer terms, for each selected service the reviewer receives a percentage of the savings attributed to its reviews.⁴ The entity deciding whether a service should be paid earns more when fewer services are paid.

CMS pairs that incentive with counterweights, also in the published record. Participant payments carry quality adjustments tied to performance measures in three categories: process quality, which explicitly includes the number of non-affirmations and favorable appeal decisions — reversals count against the reviewer; provider and beneficiary experience, including timeliness and clarity of determinations; and clinical quality outcomes, including use of alternative services and evidence of ongoing urgent clinical need.⁴ CMS also commits to monitoring downstream impacts on quality and outcomes.¹Whether those adjustments are large enough to discipline the savings incentive is not something the published documents answer; the percentage shares themselves are not in the notice.

The boundaries are explicit. The model excludes inpatient-only services, emergency services, and services that would pose a substantial risk to patients if substantially delayed.⁴ It does not apply to Medicare Advantage. It changes no coverage rule, no payment rate, and no documentation standard — the criteria are the same NCDs and LCDs that governed these services before, applied earlier.¹ And it does not touch appeal rights: a provider facing a denial after non-affirmation holds the same administrative appeal ladder as for any other Medicare FFS denial, on top of unlimited prior-authorization resubmissions and peer-to-peer review before the service is ever furnished.¹

The incentive arithmetic is unpublished. The notice establishes that reviewers keep a share of averted spending but not what share, per service or overall, nor how large the quality adjustments can be relative to it. Until CMS publishes model-participant agreements or evaluation data, the strength of the counterweights is unverifiable from the public record.

The technology itself is undescribed.The documents establish what the AI may not do — issue non-payment recommendations without a licensed clinician — but not how its outputs are validated, audited, or disclosed. The model's stated purpose includes testing “the speed and accuracy of new technology-assisted decision-making,”¹ which implies measurement CMS has not yet made public.

The list can move in both directions. Two services were delayed within the first hundred days, the LCD lists needed a correction, and CMS reserves the right to add services in future years.¹ For a practice in the six states, the operative list is whatever the docket says this quarter — not what the launch coverage said in January.