On December 12, 2025, weeks before Medicare's new AI prior-authorization pilot went live, the company chosen to run its reviews in Ohio — Innovaccer — wrote to the Centers for Medicare & Medicaid Services that at launch it intended to "auto-affirm all prior authorization requests" (approve every one, automatically). Its reason:
"Given CMS's decision not to delay the model start date, auto-affirming is the only path available that avoids creating a backlog of unprocessed prior authorizations and claims while we finalize, validate, and deploy the full rules-based solution."
Set that letter beside a year-to-date tally from the same records, dated March 30, 2026. The company reviewing requests in Washington had turned down 3,233 of the 6,096 it decided — 53 percent. Its counterpart in Arizona had turned down 2,711 of 14,301, or 19 percent. One program, three very different relationships with the word no.
You have probably been told that, software or no software, a person makes the final call on your care. This pilot writes that promise into its rules: no denial without a clinician. I think it is the right rule. What it did not deliver was consistency, and if you live in one of the six states where it runs (or look after someone who does), the reasons matter.
One rulebook, two states, a 34-point gap
On July 28, 2026, I gave this pilot one paragraph, which ended by calling the arrangement "the states' safeguard, bolted to an incentive the states never addressed." The safeguard is the rule several states wrote into law in 2026, in varying forms — Washington's lets health plans use AI to approve requests under certain conditions, "but a licensed provider must review them to deny care based on a determination of medical necessity." The incentive is the pay. Now we can count what the pairing produces.
The pilot, WISeR (the Wasteful and Inappropriate Service Reduction model), runs from January 1, 2026 through December 31, 2031, and it is a real departure: in most cases, Original Medicare needs no prior approval at all, and WISeR adds a check for a list of services — prior approval or, if the doctor skips it, a records review before Medicare pays. It does not apply to people with Medicare Advantage. One technology company handles the reviews in each state: Cohere Health in Texas, Genzeon in New Jersey, Humata Health in Oklahoma, Innovaccer in Ohio, Virtix Health in Washington and Zyter in Arizona.
The refusals must come from people, as CMS's own FAQ says: "A human clinician with relevant clinical expertise for selected items and services must review every non-affirmation." (A non-affirmation is the pilot's word for a denial.)
The numbers come from roughly 1,000 pages of CMS records that the Electronic Frontier Foundation published on September 8, 2026 — after suing under the Freedom of Information Act — and specifically from a weekly report submitted on March 30 by Noridian, the company that processes Medicare claims for both Arizona and Washington. When The Seattle Times reported them on September 16, 2026, Rep. Suzan DelBene, a Washington Democrat who "has sponsored legislation to terminate the program," wrote to the paper: "It is staggering."
Three things belong beside those numbers. First, Noridian's report shows roughly one Arizona request in ten still undecided (1,522, against Washington's 31) and credits Virtix with "limited backlog accumulation." Second, a Virtix spokesperson told the paper that the company's affirmation rate "has trended more positively toward a rate closer to 70%." Third, as the paper noted, "Based on available records, it's unclear how many of Virtix's denials were later overturned."
Were the two states working from different lists? No. They share Noridian and a list of services — skin substitutes, reviewed in some other states, are "currently not subject to prior authorization under WISeR" in either. Whatever explains the gap, it is not a different menu.
So where does variation live, in a system with a clinician behind every no? After Noridian's February 9 report relayed an Arizona provider's concern about "potentially incorrect medical non-affirmations" caused by Zyter's AI, a CMS official passed along Zyter's reply on February 12, 2026:
"This was not a software issue. The platform does not autonomously issue non-affirmations. Instead, we identified variability in the application of coverage criteria by one physician reviewer."
Zyter said it reviewed the affected cases and retrained the clinical entity involved, leadership included. The safeguard worked as its defenders hope — a complaint, a look, a correction. By Zyter's own account, it is also where the variability came from, inside the vendor with the lower denial rate. A clinician on every denial guarantees a human decision, not a uniform one.
Who gets paid when the answer is no?
If a human is in the loop, what does the loop reward? The notice that launched the pilot says its contractors are compensated based on "a share of averted expenditures." CMS sets a payment rate of 25 percent of a regional benchmark (roughly, the service's historical cost in that region) for each payable non-affirmation, before adjustments. As EFF's Lena Cohen puts it, "low quality scores reduce payments by only 5-10%," and vendors "are paid for requests that they deny (though not for denials reversed on appeal)."
CMS's own actuaries flagged the incentive before launch. Their memo of June 23, 2025, in the same release, reads (the ICIP is the Innovation Center's investment plan):
"The participants are paid based on the claims denied. The ICIP mentions a quality adjustment and while the determination of this adjustment has not yet been fully specified, it appears as though it would not significantly change the portion of savings the participants will receive. As a result, model participants will have an incentive to deny as many claims as possible."
And the clock? It runs one way. When CMS found Washington's vendor out of compliance with the program's 72-hour turnaround, it put the company on a corrective action plan, KUOW reported in June 2026; CMS said prior-authorization turnaround had since improved to 1.7 days. (Virtix had told Healthcare Dive a day earlier that it had received no such plan.)
The doctor's deadline works differently. In the pilot's pre-payment reviews, a provider who does not answer a records request within 45 days has the claim denied, per CMS's operational guide. No rule in CMS's guides turns a vendor's missed deadline into a yes. The vendor risks a plan and a pay cut; the patient waits.
Couldn't the difference just be the patients?
Partly, perhaps. Start with who is asking. Arizonans on traditional Medicare used WISeR's services at nearly twice Washington's rate in 2024 (43 per 1,000 beneficiaries against 24, per KFF), so the states were not starting from the same place — though that alone does not say which way it pushes a denial rate. Noridian read its own numbers cautiously: Arizona's strong approval trend "may reflect submission type, provider familiarity, or workflow maturity," and Washington's higher denial trend "warrants continued monitoring and targeted provider education efforts." It passed no verdict on either vendor.
The human layer also does real work. In Texas, the Washington Post reported, a physician's review lifted approvals from about 62 percent to 84 percent — though Georgetown's Center on Health Insurance Reforms, which relays the figures, cautions that they "may not be representative of overall model performance."
CMS's design answers the incentive several ways: a clinician on every no; payment only "once per beneficiary, regardless of the number of resubmissions, which incentivizes an accurate determination at the time of the first request"; nothing for noes reversed on appeal; audits that can cut a vendor's pay or remove it from the pilot; and corrective action when timeliness slips. The Zyter episode is its best exhibit.
The deepest objection comes from the Manhattan Institute's Chris Pope, who backs bringing prior authorization to traditional Medicare ("There's precedent for this working") but writes here about it in general: "'Medical necessity' is not a clear-cut matter but an inherently debatable standard for health-care procedures, which exist on a spectrum of cost-effectiveness." If so, reviewers will disagree, and a gap between two states is simply the nature of the thing. (Pope also says prior authorization "certainly requires oversight.")
I accept the premise. It just moves the question: if disagreement is built into medicine, who absorbs it — the patient's calendar, or the payer's? WISeR answers that in its clocks and its pay.
In Germany, the insurer is the one watching the calendar
What if the clock ran the other way? Germany's statutory health insurance offers one working answer. (To be clear, this compares designs; nothing here shows German patients are refused less often.)
Under section 13(3a) of the Fifth Book of the Social Code, a sickness fund must decide an application quickly — within three weeks at the latest, or within five if it seeks an expert opinion (above all from the Medizinischer Dienst, the system's medical review service) and tells the patient so. Miss the deadline without notifying a sufficient reason and the benefit, in an unofficial English translation, "shall be deemed to have been approved on expiry of the deadline"; a patient who then obtains a necessary treatment can claim the cost back.
Now the catch. On May 26, 2020, Germany's Federal Social Court abandoned its own earlier case law and cut that promise back. A missed deadline no longer creates a free-standing right to the treatment. It gives the patient only a provisional right to obtain it themselves and be reimbursed — and only in good faith — while the fund stays entitled, and obliged, to decide. What survived is the point, which the court located in the pressure it puts on the funds: after the deadline, a fund cannot argue that a treatment the patient already obtained was substantively unlawful.
Then there is the reviewer. The Medizinischer Dienst is financed by a per-member levy on the sickness funds. The service says its flat fee — about 15 euros per insured person in 2023 — does not change with how often or why it is asked, how complex the review is, or what result it reaches. It advises; the fund decides.
Side by side, as I read the two rulebooks, the difference is who carries the cost of delay and doubt. In WISeR, a late provider's claim is denied, a late vendor risks its quality score, every payable no earns a fee, and the patient waits. In Germany, a missed deadline tilts the default toward the patient, however narrowed; a refusal can be contested directly and taken to the social court free of charge; and the reviewer's fee is the same whatever it concludes.
Picture the map in 2031
WISeR's own planning papers imagine more. A June 2025 draft of the investment plan lists possible future services, including air ambulance transport (where, the planners wrote, untimely approval "could result in a delay of medically necessary, emergent care") and oncology care such as "CAR T-cell therapy, radiation oncology." CMS has indicated, the Seattle Times reports, that it may add services and "may then expand the program to other states," and Abe Sutton, who runs the Innovation Center, said on September 3, 2026 that WISeR "should serve as a model for other lines of business" — while promising to "learn from its lessons before we look to expand."
So picture 2031 — this part is my extrapolation. Two patients with the same cancer, in two WISeR states, need the same cell therapy. Each request goes to a different company, each paid a share of what a no saves, each on a clock that costs the company a plan when it slips and costs the patient time. One oncologist gets a yes on the two-day expedited track; the other gets a no, resubmits, books a peer-to-peer call, and waits. Neither company broke a rule.
Meanwhile, under CMS's participant guide, doctors who reach "a minimum affirmation threshold of 90%" — each vendor may set a lower bar — can be exempted from review. By my arithmetic, Washington's vendor affirmed 47 percent of its first-quarter decisions and Arizona's 81 percent, so I would expect far fewer Washington doctors to earn that exemption, compounding the gap without anyone deciding it should.
What the pilot's friends concede, and what its critics add
On the right, Paragon Health Institute calls WISeR "exactly what the Innovation Center should be doing", reading Medicare Advantage data as proof that prior authorization works: "88 percent of denials are not appealed and most of the reversals are providers correcting paperwork gaps." The Senate voted 46–50 along party lines on July 16, 2026 to preserve it. AEI's James Capretta writes that a fee based on avoided costs gives vendors "a strong incentive to screen out claims that fall short of the clinical criteria established for the various targeted services." He calls the pay plan perhaps "the most controversial aspect of the model"; the first quarter asks who decides what falls short.
In the center, KFF notes that "prior authorization remains one of the few tools available to insurers to manage health care utilization and spending." And Henry Bair, writing in the American Journal of Medicine Open, warns that "Even when a clinician makes the final determination, AI-assisted workflows can powerfully shape decisions," while granting that "Some critiques of WISeR understate the guardrails CMS has already specified."
On the left, Don Berwick — a former CMS administrator — and Andrea Ducas, both of the Center for American Progress, argued before launch that one of traditional Medicare's advantages is that it "almost entirely lacks the onerous preauthorization reviews that plague the program's privately administered Medicare Advantage option."
Put AEI's "strong incentive" next to the Manhattan Institute's "inherently debatable standard" and you have the whole problem in two phrases, both from right-of-center scholars. The argument is only over whether a clinician's signature is enough to manage it.
What does this mean for you?
If you or someone you help is on Original Medicare in one of the six states, here is what CMS's documents allow:
Check whether WISeR applies. It covers Original Medicare only (not Medicare Advantage), in the six states, for listed services. Your doctor's office can tell you if your procedure is one.
If the answer is no, get the reasons. You get a copy of the non-affirmation, and CMS's operational guide says the vendor must detail "all missing and/or noncompliant information that led to the decision," plus instructions for resubmitting. Ask the office for it.
Ask your doctor to resubmit, and to request a peer-to-peer review each time. There is no limit on resubmissions until a claim has been submitted and denied, and providers "may request a peer-to-peer clinical review of each resubmission." The office files these, not you.
If waiting could seriously harm you, ask your doctor to request expedited review. It is meant for delays that could seriously jeopardize your life, health or ability to regain maximum function. If the vendor agrees the risk is real, it has two days to decide; if not, the request goes on the standard clock, typically three calendar days.
Know what you can't appeal, and what you can. The non-affirmation itself "would not be appealable." If you go ahead anyway, CMS's FAQ says your physician must first give you an Advance Beneficiary Notice of Non-Coverage (the form warning that Medicare may not pay, and that you may owe the bill). If Medicare then denies the claim, you or the provider can appeal it — and the same goes for a claim denied after a records review.
Appeal on time. File by the date on your Medicare Summary Notice, within 120 calendar days of receiving it. Your odds are unknown, because WISeR's appeal results have not been published.
Get free help. Your State Health Insurance Assistance Program (SHIP) offers one-on-one counseling, including "Making an appeal or complaint," and "There is no cost to consumers to use SHIP services." Call 1-877-839-2675 to find yours.
Know who to call. In Washington, Virtix's WISeR line is 833-943-2209; in Arizona, Zyter takes WISeR questions at 202-773-1430 (9 a.m. to 4 p.m. Arizona time) or wiser@zyter.com. CMS's WISeR team is at WISeR@cms.hhs.gov. Other states' vendor contacts are in the operational guide.
The lesson, as I see it
A clinician behind every denial is a floor, and worth defending. But a floor guarantees only that a person signs each no. On the first quarter's evidence, it does not make outcomes consistent from one contractor to the next, and it does nothing about the pay or the clock.
What would help? Publish outcomes by state and vendor, including how many noes were reversed on resubmission and on appeal. Put a consequence on the payer's clock that lands in the patient's favor, as Germany does even in the narrowed form its court left standing. And pay reviewers in a way that does not move with their conclusions.
The next data point is CMS's own — Sutton said the Innovation Center plans an "early snapshot of how it's rolling out," drawn from the first six months, in the last quarter of 2026. Look for three numbers by state: the non-affirmation rate, the share reversed, and the days waited. If it carries only the first, read it as a press release.
My vote? Keep the clinician. Change the clock and the pay. A signature on every no was the right first step; it was never going to be the whole design.
If someone you care about in one of the six WISeR states is waiting on a yes from Medicare, send this to them, or to whoever keeps track of their appointments. The HAIA Foundation would rather you knew the resubmission rules before the letter arrives.