Physicians spend more than 16 hours per week on prior authorization. Forty-three requests per week, per physician, according to the AMA's 2025 survey. Most of that time is not clinical work. It is navigating IVR menus, waiting on hold, reading CPT codes aloud, and waiting for a payer rep to type what you just said into their system.
An AI voice agent handles that part. Not the clinical judgment. The phone call.
What prior authorization actually requires
Before automation makes sense, the workflow needs to be understood precisely.
A standard prior authorization call has six steps. The practice calls the payer's utilization management line. The IVR asks for the member ID, date of birth, provider NPI, and the reason for the call. Hold times average 22 minutes. When a rep answers, they need the CPT code, the diagnosis code, the clinical rationale, and the ordering physician's NPI. The rep generates an authorization number. Someone on your staff writes it down and enters it into your practice management system.
Step one through five is structured, repeatable, and requires no clinical judgment. It requires accuracy, patience, and time.
That is the part an AI agent automates.
What the AI voice agent does
The agent places the outbound call to the payer. It navigates the IVR without a human listening. It waits on hold. When a rep answers, it reads the member ID, the CPT codes, the diagnosis codes, and the NPI from the patient's record in your EHR.
It captures the authorization number and writes it back automatically. No transcription. No sticky note. No entry error.
The 2024 CAQH Index Report puts the manual cost of a single prior authorization at $10.97 for providers. Automated processing drops that to $5.79. For a practice submitting 40 PAs per week, that is over $10,500 per year in direct savings. Staff time is additional.
Practices using automation also report 30% fewer initial denials. Not because the agent argues better. Because submissions are complete and accurate the first time. Manual calls introduce errors. The agent reads exactly what is in the record.
What it cannot touch
Clinical escalations stay human.
When a payer routes to a clinical reviewer for complex cases, the agent detects the escalation. It either holds for a staff member to join the call or flags the case for a callback. It does not attempt to defend clinical decisions, provide rationale not in the record, or negotiate with a medical director.
82% of prior authorization appeals succeed when filed, according to the AMA. Fewer than 11% of denied requests ever are filed. That gap is a clinical argumentation problem, not a phone call problem. It is a physician's work or a trained appeals coordinator's work. The agent does not touch it.
The split is straightforward. If the step requires reading a value from a record, the agent handles it. If the step requires judgment, a human handles it.
The failure mode question
Every prior authorization automation has a failure mode worth understanding before you build.
If the agent cannot reach the payer, the call needs to be flagged immediately and retried on a schedule. If the agent misreads a code, the denial lands quietly and the staff member who would have caught it was not on the call. If the EHR write-back fails, the authorization number is gone.
The first question before building any PA automation is not what can we automate. It is what happens to patient care if this breaks at 9pm and no one notices until morning. The answer determines how tight the error handling needs to be and what monitoring looks like after go-live.
In a clinical workflow, the cost of a broken automation is not a missed invoice. It is a delayed procedure.
Why phone calls are not going away
CMS mandates Prior Authorization Support FHIR APIs for Medicare Advantage, Medicaid, and CHIP plans by January 1, 2027. Commercial payers are exempt from that mandate.
Commercial insurance covers the majority of privately insured patients in the US. Phone-based prior authorization remains the dominant modality for most payers for at least the next two to three years. Building on that assumption is safe. Building on the assumption that APIs will replace phone calls by next year is not.
The PA volume is also growing. Medicare Advantage alone saw 52.8 million prior authorization requests in 2024, up 42% since 2019. The problem is not shrinking while the industry waits for FHIR.
When this is not the right fit
If your practice submits fewer than 20 prior authorizations per week, the build cost exceeds the time savings. A trained staff member with a good checklist handles it.
If your payer mix is primarily Medicare Advantage plans already on electronic PA pathways, the phone call volume is lower and the ROI calculation changes. Audit your payer mix before deciding.
If your EHR does not have an accessible API or a supported integration method, the write-back requires custom development. That changes the scope. We will tell you this before work starts, not after.
What a realistic build looks like
A single-payer prior authorization automation takes two to three weeks to build and test. One payer's IVR mapped, one EHR integration, one authorization number write-back. Multi-payer builds add time proportional to the number of IVR trees being mapped.
Each payer's IVR is different. United and Cigna are not the same call. The agent needs to be trained on each one.
Testing happens against real call flows before anything goes live. We do not deploy against a payer that has not been walked through end to end in a test environment.
After go-live, payer IVR trees change without notice. The monitoring catches it. We fix it. You are not left with a broken workflow you did not build and cannot see into.
Get your free automation scope and we will map your payer mix, your EHR setup, and the PA volume that makes automation worth building. We will tell you if it does not.
Sources: AMA 2025 Prior Authorization Survey · CAQH 2024 Index Report · KFF Medicare Advantage Prior Authorization Data · AMA on PA Appeals