Prior Authorization Is Your Most Automatable Process. It Is Also the One Where Speed Hurts Patients

The one process everybody agrees is broken

Ask a practice administrator which process they would automate first and prior authorization comes up before you finish the sentence. It is the right instinct. The AMA's physician survey puts the load at an average of 39 authorizations per physician per week and about 13 hours of clinician and staff time to move them. The 2024 CAQH Index found only 35 percent of medical prior authorizations run fully electronically through the standard transaction, which means most of that volume is still portal logins, hold music and fax confirmations.

So the pitch writes itself, and about forty vendors have written it. What none of them will tell you is that prior authorization is not one process. It is two, they have opposite risk profiles, and the automation case is strong for one of them and genuinely dangerous for the other.

Two processes wearing one name

The first process is the submission packet and the status chase. Which payer, which form, which attachments, which portal, has anyone heard back, is it time to call again. This is rules-driven work. It does not require clinical judgment, it does not vary much within a payer, and the people doing it are expensive and bored. Automate it.

The second process is the medical necessity argument. Whether this patient, with this history, on this failed prior therapy, warrants this drug or this imaging study, and how that case is framed to a payer that has an economic reason to say no. That is clinical reasoning written down. It looks automatable because it arrives as a form, and forms are what software eats.

Almost every group that gets burned here made the same move. They automated submission without fixing which cases get submitted and in what shape. The result is that the same weak packets now go out faster and come back denied faster. In an accounts payable process that is an annoyance. Here the output of the process is a patient waiting on a treatment, and 29 percent of physicians in the AMA survey reported a prior authorization delay that led to a serious adverse event for someone in their care. You do not want to be more efficient at producing that.

The AMA's own position on automated coverage decisions is a fair standard to hold your own operation to, not just your payers': AI should support physician judgment, not substitute for it. If a payer using an algorithm to deny without qualified clinical review is unacceptable, a practice using one to submit without clinical review is the same argument pointed the other way.

Count your volume before you shop

Here is the diagnostic that decides the purchase, and it takes a week.

Pull twelve months of authorizations. Sort them two ways: by payer, and by service line. Then look at how flat the distribution is.

In most groups we have looked at, it is not flat at all. A handful of payers and two or three service lines produce the clear majority of the volume, usually because of one drug class, one imaging modality, or one procedure that a particular plan has decided to gate. That concentration is the whole buying decision. If sixty percent of your authorizations run through three payers, the thing that fixes your problem is three payer-specific workflows, not a platform priced across every payer you have ever billed.

Vendors will not run this analysis for you, because the answer is frequently that you need about a fifth of what they sell. This is the same buy-side question we walk through in how to evaluate an AI vendor: does the product address the shape of your volume, or the shape of their category.

The second cut of the same data is more uncomfortable. For each concentrated group, ask what your approval rate is. A service line running at 97 percent approval is not an automation candidate, it is a negotiation candidate or a gold-carding candidate. You are staffing a process whose outcome is a foregone conclusion. Automating it makes a pointless activity cheap enough to keep forever, which is the specific way automation entrenches waste rather than removing it. We have made this argument in other contexts too: when AI is the wrong answer to an operations problem is usually when the honest fix is to stop doing the work.

Some of what you are being sold arrives as regulation

This is the part of the market that buyers keep paying for twice.

CMS-0057-F is now partly live. Since January 1, 2026, impacted payers, meaning Medicare Advantage organizations, state Medicaid and CHIP programs and their managed care plans, and qualified health plan issuers on the federally facilitated exchanges, have had to decide expedited requests within 72 hours and standard requests within seven calendar days, and to give a specific reason for every denial. The Prior Authorization API requirement lands January 1, 2027.

Read that as a buyer. A meaningful share of what a prior authorization platform demos as proprietary value, faster turnaround, structured denial reasons, an electronic path to the payer, is scheduled to show up in your covered payer mix inside sixteen months because the rule requires it. Separately, more than fifty insurers pledged in 2025 to reduce the services they gate at all, with several naming double-digit cuts to their code lists and new gold-card programs. The AMA found only about 11 percent of the burden actually removed a year in, so nobody should plan around the pledge. But you should absolutely refuse to sign a five-year contract that prices the pledge as if it will never happen.

The question to put to a vendor, in the demo, is simple. Which of these capabilities survives CMS-0057-F, and what does your product do that a compliant payer API will not do for free? A good answer exists. It is about the workflow on your side of the wire, the packet assembly, the denial triage, the appeal routing, none of which the rule touches. A vendor who cannot separate their product from the regulation is selling you a calendar.

What good looks like

The groups that get this right end up with an automation footprint that is narrower than they expected and stickier than the platform would have been.

Status chasing goes first and goes fully. Nobody should be logging into a portal to ask a question a transaction can ask. Packet assembly follows: pulling the same twelve artifacts for the same payer every time is a retrieval problem, and retrieval is what this technology is actually good at. Denial triage is where the returns concentrate, because denials sort cleanly into reasons that are administrative, meaning a missing field or a wrong code that should be corrected and resubmitted automatically, and reasons that are clinical, meaning a human needs to read the chart before anything goes back. That routing decision is worth more than the submission speed everyone shops for, and it is the same structural insight behind workflow orchestration: the value is usually in the handoff, not the task.

What stays human is the medical necessity narrative and the peer-to-peer. Not because software could never draft it, but because the group that keeps a clinician's eyes on that step is the group that does not find out in a deposition that its automation had a house style.

We keep coming back to one framing with clients on this. Prior authorization does not need to be faster so much as it needs to be smaller. Speed applied to the wrong queue just gets you to the wrong answer sooner. A shop that cuts its authorization volume by a third through payer negotiation, gold carding and scope cleanup beats a shop that processes the same volume twice as fast, and it costs less than the platform. Call it the only case where the best automation project is the one you talk yourself out of, which is a hard thing to sell and an easy thing to prove.

If your authorization volume is concentrated, your approval rates are high in the places it hurts, and you are being quoted a platform price for a payer-specific problem, that is worth an hour before it is worth a signature. We do this work from the buying side, and we do not sell the software. Bring us the volume analysis and the vendor quote, and we will tell you which parts of the process should be automated, which should be renegotiated, and which should simply stop.