AI Agents for Medical Practices: Prior Authorization and Eligibility

An AI agent for a medical practice automates prior authorization submissions, insurance eligibility checks, and referral status tracking, the three tasks that eat the most non-clinical staff time in a practice. Suvysoft builds these as custom agents wired into the practice management system and payer portals already in use, typically $6,000 to $15,000 to build and $300 to $800 a month to run.

Most guides to this topic read like a feature list for prior authorization software. What they skip is the thing that should decide how a practice builds or buys one right now: a federal rule with a January 2027 deadline is about to change the plumbing these agents run on, and an agent built to scrape a payer portal today may need rework the day that deadline lands.

What does an AI agent for a medical practice actually automate?

Three jobs sit in the same category because they share a pattern: pull structured data out of a payer system, fill in what a person would otherwise type by hand, and flag anything that does not match a clean rule for a staff member to review. Prior authorization submission and status checks are the first. Insurance eligibility and benefits verification before a visit is the second. Referral status tracking between a referring practice and a specialist is the third.

None of the three requires clinical judgment, which is exactly why they are the right first target for automation and the wrong place to stop asking questions about it. A well-scoped agent gathers the documentation a payer requires, submits through the payer's portal or API, checks status on a schedule instead of a staff member calling in, and routes anything denied or flagged back to a person with the relevant chart data already attached.

What does prior authorization actually cost a practice right now?

The honest number is bigger than most vendor pages lead with. Physicians complete an average of 40 prior authorizations a week, consuming roughly 13 hours of physician and staff time in that same week, according to the American Medical Association's 2026 prior authorization physician survey. The same survey found 95% of physicians report prior authorization delays access to necessary care, and 40% of practices now employ staff whose job is exclusively prior authorization work.

The per-transaction cost varies by specialty more than most estimates acknowledge. CAQH's 2023 Index Provider Specialty Issue Brief puts a manual prior authorization at $7.60 for a generalist practice and $15.12 for a specialist practice, against $4.47 and $6.61 respectively when the same transaction runs electronically, a 41% to 56% cost reduction depending on specialty. None of that counts the revenue sitting behind a delayed procedure while the paperwork is pending.

How does the 2027 CMS interoperability rule change what a prior authorization agent should do?

This is the part most articles on this topic skip entirely, and it is the reason the build decision matters more this year than last. The CMS Interoperability and Prior Authorization Final Rule, CMS-0057-F, already requires impacted payers, Medicare Advantage, Medicaid, CHIP, and ACA marketplace plans, to decide urgent requests within 72 hours and standard requests within seven calendar days, and to publish a specific reason for every denial, as of January 1, 2026.

The bigger change lands January 1, 2027, when those same payers must expose four FHIR-based APIs, including a dedicated Prior Authorization API that lets a provider system query requirements and submit a request programmatically instead of through a payer's web portal. A prior authorization agent built today to log into a portal and fill a web form is solving 2025's problem. An agent built to query a payer's API for documentation requirements before submission, and structured to swap the portal step for an API call once a given payer turns theirs on, is solving the problem that actually exists in fifteen months.

Practices evaluating a build now should ask any vendor or development partner one direct question: does this agent's architecture assume portal automation permanently, or is the payer-facing layer built to be swapped for an API call without a full rebuild? The answer determines whether this year's build cost gets used twice.

What does eligibility verification cost when a person does it by hand?

Insurance eligibility and benefits verification is the check that happens before a prior authorization is even needed, and it carries its own real cost. CAQH's index data, presented in the organization's own 2023 Index webinar materials, puts a fully manual eligibility check at $7.97 per transaction against $2.18 for a fully electronic one, a difference that compounds fast across a full appointment schedule.

A practice running 120 appointments a week with insurance coverage on 80% of them is running roughly 96 manual-equivalent checks weekly if none of that is automated, and at 15 minutes per call that is 24 hours of staff time before a single prior authorization has even started. An agent that runs eligibility checks against a payer's real-time API the moment an appointment is booked removes that step from the front desk's morning entirely, and flags a lapsed or changed plan before the patient is in the chair rather than after the claim is denied.

Where does referral tracking fit, and where does it stop?

A referral agent watches the gap between a referral being written and a specialist visit being completed, the step that most commonly goes unmonitored because no single system owns it end to end. It checks whether the specialist's office has scheduled the patient, follows up automatically if nothing has happened within a set window, and surfaces the referral to staff once it has stalled long enough to need a phone call.

What it does not do is decide where a referral should go or override a physician's specialist choice. The agent's job stops at visibility and follow-up. A referral that needs a different specialist because of network status, patient preference, or clinical judgment is a decision for the referring physician or a care coordinator, not the automation layer.

What should stay with clinical staff, never the agent?

The agent should never be the one that decides a denial is final, that a symptom does not warrant a call back, or that a patient's authorization gap means a visit gets postponed without a person reviewing the chart first. Prior authorization, eligibility, and referral tracking are all administrative workflows sitting adjacent to clinical decisions, and the line between them has to stay explicit in how the agent is built, not just in a policy document nobody reads after go-live.

A practice's Business Associate Agreement with any AI vendor or development partner needs to name exactly what the agent can read, write, and act on inside the EHR, because every one of these three workflows touches protected health information from the first API call. That agreement, along with a written record of what the agent is and is not authorized to do, is the same documentation a payer or accreditor will ask for if a denial or an eligibility error is ever disputed.

What does it cost to build and run this kind of agent?

Pricing follows the same shape as any single-workflow AI agent: a build cost to connect it to the systems a practice already runs, then a monthly cost tied to claim and appointment volume.

Cost moves with how many payer systems the agent connects to, whether the practice's EHR exposes a usable API or requires portal-level automation for now, and how many locations share the same build. A single-location practice on a modern cloud EHR sits at the low end of each range. A multi-location group connecting to a dozen different payer portals, several without a current API, sits toward the top.

An eligibility-only build wired to a practice's scheduling system typically takes two to three weeks. Adding prior authorization pushes that to four to six weeks, mostly spent mapping each payer's documentation requirements rather than building the underlying agent logic. Adding referral tracking on top of both usually adds another two to three weeks. Suvysoft's custom agents work scopes each of these against the practice management system and payer mix already in place, with the API-versus-portal question above built into the architecture decision from the start rather than revisited after the 2027 deadline forces it. See the full range of agentic AI work Suvysoft builds, including how initial AI setup scoping works before any build cost is quoted, and recent case studies for how similar engagements were scoped.

Frequently asked questions

What does an AI agent for prior authorization actually do?

It gathers the documentation a payer requires for a given procedure or medication, submits the request through the payer's portal or API, checks status on a set schedule instead of a staff member calling in, and routes anything denied or flagged for additional information back to a person with the relevant chart data already attached. It does not decide whether a denial is final.

How much staff time does prior authorization take without an agent?

The American Medical Association's 2026 survey found physicians complete an average of 40 prior authorizations a week, consuming roughly 13 hours of combined physician and staff time in that same week. Forty percent of practices now employ staff whose job is exclusively prior authorization. That number scales roughly with a practice's procedure and prescription volume.

What happens to prior authorization once the CMS API rule takes effect in 2027?

Starting January 1, 2027, payers covered by CMS-0057-F must expose a Prior Authorization FHIR API, letting a practice's system query requirements and submit requests programmatically instead of through a web portal. An agent built with that swap in mind avoids a rebuild when a given payer turns its API on; one built assuming permanent portal automation will likely need reworking.

Can an AI agent make the final call on a denied prior authorization?

No. The agent can draft an appeal, gather supporting documentation, and flag the denial for review, but the decision to appeal, accept, or escalate a denial has to go through clinical or billing staff. Automating the paperwork around a denial is a time savings; automating the judgment call on it creates liability the practice's agreement with any vendor needs to explicitly rule out.

How much does an AI agent for a medical practice cost?

An eligibility-only agent typically runs $3,000 to $6,000 to build and $150 to $300 a month to run. Adding prior authorization pushes that to $6,000 to $10,000 to build and $300 to $500 a month. A full build covering eligibility, prior authorization, and referral tracking across multiple payer systems runs $10,000 to $15,000 or more to build, with $500 to $800 or more a month depending on volume.

Want to know what an eligibility, prior authorization, or referral agent would actually cost for your practice's EHR and payer mix? Talk to us about scoping one against your real workflow before anything goes live.