On January 1, a rule that was supposed to make prior authorization less painful took effect. CMS now requires most federally regulated plans — Medicare Advantage, Medicaid, CHIP, and the Exchange plans — to decide standard prior-auth requests within seven calendar days instead of fourteen, and expedited ones within 72 hours.
Faster decisions. Good news, on paper.
Here is what actually happened. Prior-authorization denials are running about 31% higher year over year in 2026. The clock got shorter; the “no” got more frequent.
That is the trap of this whole moment, and it is worth naming plainly. A deadline forces a decision. It does not force a yes. When you give an understaffed utilization-review operation less time and the same backlog, the safest output under pressure is a denial the provider can appeal later. The rule optimized the wrong variable.
Meanwhile CMS is running its own experiment. The WISeR model went live January 1 in six states — New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington — and for the first time it puts prior authorization on Original Medicare for a set of outpatient services, with technology vendors doing the reviewing. Whatever you make of it, the direction is unmistakable: more automated review, in more places, on more services.
So every revenue-cycle leader is now shopping for AI to fight back. Roughly 63% of health systems have already put AI somewhere in the revenue cycle, and about 80% are piloting generative tools for it.
That instinct is right and the sequencing is usually wrong. AI applied to a clean prior-auth process compounds. AI applied to a messy one just generates denials — or appeals — faster. If you do not know which of your denials were avoidable, a model will simply help you lose at speed.
From the Playbook
The Revenue Cycle playbook in our series opens with one unglamorous move, and it is the one to make this month: build a denial taxonomy before you buy anything.
Pull ninety days of denials and sort them into exactly three buckets. One, avoidable and ours — missing documentation, wrong code, no auth on file. Two, avoidable and theirs — the payer denied something that met criteria. Three, not covered, full stop. Most teams have never done this, and the number that matters falls out immediately: your avoidable-and-ours rate. That is the only bucket AI can fix cheaply, and it is usually the biggest. Automate that first — clean-claim checks, documentation prompts, auto-attached auth numbers — and you will recover more than any denial-appeal bot wins back downstream. Fix the intake, and the seven-day clock stops working against you.
One number to anchor it: if a third of your denials are avoidable-and-ours, that is a third of your appeal labor, your rework, and your delayed cash that never had to exist.
What we are watching
The big systems are already booking the savings. HCA expects roughly $400 million in AI-driven cost reductions this year, a share of it from automating revenue management, and UnitedHealth projects close to $1 billion across claims, billing, and documentation. Those are not pilots. When operators that size move real money onto the line item, the question stops being whether the tooling works and becomes whether your process is clean enough to point it at.
One thing to try this week
Pull your last ninety days of denials and tag each one avoidable-ours, avoidable-theirs, or not-covered. Do not fix anything yet. Just get the three numbers. You cannot automate your way out of a denial rate you have never sorted.
The Operations Edge is published by the Healthcare AI Institute — practical, vendor-neutral playbooks for the people who run healthcare operations. Fix the operation before you automate it.
The full series and toolkits: Book 0 — The Primary Care Operations Playbook. Book 1 — AI for Healthcare Scheduling & Patient Access. Toolkits and templates on Gumroad.