Healthcare

Is prior authorization a phone problem?

Image

Chetan RK, Product Manager Agentic AI

Picture a prior authorization coordinator at nine on a Tuesday morning. A queue of open requests. A handful already past the date the payer should have decided.

She dials the first one. Press 1 for English. Press 2 if you’re calling from a provider office. Press 3 for prior authorization. Press 2 to check a status.

Then the recording she knows by heart. The system can’t retrieve the status automatically. Please hold. And while she holds, a friendly reminder that she can check status anytime on the portal. She already did. That’s why she’s calling.

Multiply that by every open request, every payer, every day. That’s where prior authorization actually lives. On hold.

Everything after the request

When people talk about prior auth, they talk about the request: the forms, the clinical criteria, the portal. Those matter. But the time goes somewhere else.

The AMA’s latest physician survey puts it at 13 hours a week of physician and staff time, across 40 requests per physician. CAQH puts a single manual request, by phone, fax or email, at 24 minutes of provider and staff time. And the cost isn’t only time. 95 percent of physicians say prior auth delays the care their patients need, and 79 percent say it can lead patients to abandon treatment. That first number has been above 90 percent for three surveys running.

The work after the request is a short list of questions. Has it been decided? If it was denied, why? What does the payer need? Where does the appeal go, and by when? Most of those answers live with one person: a representative at the payer, at the end of a phone line.

The rules got faster. The phone didn’t.

2026 was supposed to fix some of this, and in places it has. Since January, a CMS rule requires Medicare Advantage, Medicaid and CHIP plans to decide standard requests within seven calendar days and urgent ones within 72 hours, and to give a specific reason for every denial. In March, those payers published their first prior authorization metrics. By 2027 they must offer prior authorization through FHIR APIs, and health plans have pledged real time answers for 80 percent of electronic approvals by then.

The first published numbers show why it matters. In KFF’s analysis of those reports, 12 percent of standard Medicare Advantage requests were denied.

All good. None of it picks up the phone.

A faster clock means a missed decision is visible on day eight. Someone still has to chase it. A required reason still has to be read, sorted and acted on before the appeal window closes. The clocks don’t cover employer plans at all. Benefit managers run their own lines and their own menus. And some requests simply go quiet.

Four menus deep

In a demo we recorded, reaching the status of one request took four menus. Then a dead end: the system couldn’t retrieve the status. Then the hold.

This is the maze our CTO, Sriram Chakravarthy, wrote about in Teaching AI agents to navigate the maze, seen from the other side. There, the AI walks a company’s menus on a customer’s behalf. Here, it walks a payer’s menus on a health system’s behalf. The stakes are a patient’s MRI.

A denial is a deadline

A denial isn’t the end of a request. It’s the start of a clock.

Appeal windows run from the date of the notice, and the notice may still arrive by mail. The reason in it may be one line. Meanwhile the work queue still says pending, and the date of service keeps coming.

So the questions on a denial call are specific. What exactly was the reason? What would the payer need to see? Is an expedited appeal available? Where does it go, and what is the last day? Ask all of them on the first call and the appeal can start that afternoon. Miss one and it’s another call, and another hold.

Meet Nora

Nora is Avaamo’s prior authorization agent. She works the insurance phone lines.

She starts in Epic. Nora reads the pending authorization work queue your team already uses, and calls first for the requests that are urgent or furthest past the payer’s decision date. Your rules set the order.

She works the line. She navigates the menus, waits out the hold, and when a representative picks up, she says she is an automated assistant. Every time.

She reads it back. Status, denial reason, appeal deadline, where the appeal goes, the representative’s name, the call reference number. Each answer is read back and confirmed before it is recorded.

She writes it where your team works. The decision, the reason and the reference land in the existing Epic fields, with a call summary in the comments, and the request moves to the right work queue.

She sorts the denial. Avaamo PA Resolve reads the reason as stated, matches it to a denial type, picks the next step and sets the deadline. A documentation gap becomes a written appeal, due on a specific date, with the notes it needs.

She follows through. The appeal is drafted for a physician to review and sign. Nora calls to confirm the payer received it, checks status until there is a decision, and texts the patient when there is an approval.

The catch

The moment we keep coming back to in our recorded demo call isn’t the menu or the hold. It’s a fax number.

The representative reads out the fax for appeals, stumbles, and corrects the last digits. A person deep into a day of calls might write down either version. Nora heard two different endings, asked for the full number again, digit by digit, and read it back before recording it.

That’s the job. Not sounding human. Getting it right.

Saying what she is, getting the answer right, and leaving a record anyone can check. That is what earns an agent a place on a payer line.

Where this fits

Nora doesn’t replace the electronic routes. When a payer’s answer reaches Epic through them, the request leaves the queue and Nora never dials. She doesn’t fax, and she doesn’t log into payer portals. She takes the calls that are left.

She doesn’t replace your team either. She takes the hold time, so the people who know your patients and your payers spend their day on the exceptions: the peer to peer, the difficult appeal, the patient who needs a call from a person.

Start the way we start with every agent. One queue, your own baseline, and the next queue when you see it work.

Prior auth will live on the phone for a while yet. Your team doesn’t have to live there with it.