Medicare's AI prior authorization pilot: what to track now
For years, prior authorization was something you dealt with in Medicare Advantage and commercial plans. Traditional Medicare mostly left it alone. In parts of the country, that has changed.
CMS’s WISeR model adds prior authorization to some services in traditional Medicare. Outside companies run the reviews, with AI helping them decide. It’s a pilot, and it’s still early.
If your health system is in a WISeR state, you’re already living with it. If you aren’t, you may be later. Either way, the early months are worth watching closely.
What the early months show
There isn’t much public data yet. Most of what we know comes from health systems and reporters.
The clearest report so far is from Washington state. UW Medicine told KUOW that approvals that used to take one to three days were taking 15 to 20 days under the program. That’s one system’s experience, and it may improve as the program settles in. But it’s the kind of delay patients feel.
The program’s future is also less certain than it looked. In May, GAO ruled that the WISeR notice counts as a “rule” under the Congressional Review Act. That gives Congress a path to try to repeal it. Whether that happens is a political question. I wouldn’t plan around either outcome.
Some context helps. For Medicare Advantage, Medicaid and federal-exchange plans, CMS now requires prior authorization decisions within 72 hours for urgent requests and 7 calendar days for standard ones. Those deadlines are for other programs. They still make a useful yardstick, and a 15-day wait looks long next to them.
What to track on your side
You can’t control how a vendor reviews a request. You can know your own experience in detail.
While public data is thin, your own data is the most useful thing you have. It helps you run your operations. It also gives you facts for any conversation with CMS, your state hospital association, or your members of Congress.
Start with a simple log of every WISeR request. For each one, capture:
- The service and the ordering clinician.
- The date you submitted it and the date you got a decision.
- The outcome: approved, denied, or sent back for more information.
- What happened next: resubmitted, appealed, or dropped.
Most of this already lives somewhere in your systems. The work is pulling it into one place and keeping it current.
Then look at a few numbers each month. How long decisions take, compared with the same services before the program. How often requests are denied or sent back, and how many of those are later approved.
Add one more: how many procedures were rescheduled or cancelled while you waited. That number matters most. A delayed procedure is a patient waiting, and often an empty slot on your schedule. If you can count those, you can show what the program costs in practice.
Track staff time too. Prior authorization work lands on someone’s desk. If a new program adds hours to that work, put a number on it.
Keep the log even if the numbers look fine. A clean record of fast approvals is useful too. It tells you where the program works, and it gives you a baseline if things change.
Where AI fits on your side
There’s an irony here. The review side is using AI, and in many health systems, the request side is still built by hand.
AI agents can help with the prep. An agent can gather the notes, orders and test results a request needs, check them against the coverage criteria, and draft the packet for a person to review. That can cut the back-and-forth that comes from a missing document.
An agent can also keep your log for you. It can match each request to its decision, flag the ones that have waited too long, and put them in front of someone who can follow up.
People still make the calls that matter. A clinician decides what the patient needs. Someone on your team reviews what goes out under your name. When a denial looks wrong, a person decides whether to appeal. The agent makes the work faster. It doesn’t make the decision.
This week, ask your utilization management lead one question: for the services WISeR covers, how long did approvals take last month, and how many procedures moved because of it? If nobody can answer yet, start the log.
Discussion