Automation & Workflows
What Can AI Do for Prior Authorization, and What Changes in 2027?
AI can do most of the preparation for a prior authorization: check whether a service needs one, gather the records and codes the payer asks for, fill out the request, track its status, and flag a denial with its reason. People keep the clinical judgment, the peer-to-peer calls, and the appeals. Federal rules also shorten payer decision times and require payers to offer prior authorization APIs by 2027, which makes the automation easier to connect.
How big is the prior authorization burden?
Big enough that practices hire for it. In the AMA's 2025 prior authorization physician survey, practices reported completing an average of 40 prior authorizations per physician each week, physicians and their staff reported spending 13 hours a week on them, and 40% of physicians said they have staff who work only on prior authorizations. In the same survey, 95% of physicians said prior authorization delays care at least sometimes, and 26% said it had led to a serious adverse event for a patient in their care.
The 2024 CAQH Index put a manual prior authorization at $12.88 and 24 minutes of provider staff time, against $5.38 and 10 minutes for an electronic one.
Which parts of prior authorization can AI handle?
Most of the steps, sorted into the lanes from which tasks AI can fully automate:
| Step | Who does it |
|---|---|
| Check whether the service needs authorization for this plan | AI, using the payer's rules |
| Gather notes, codes, imaging, and history the payer requires | AI |
| Draft the request or letter of medical necessity | AI drafts, a person approves |
| Submit through the portal, API, or fax | A person submits, or AI after approval |
| Track status and catch requests for more information | AI, with alerts |
| Peer-to-peer review with the payer's physician | The treating physician |
| Appeal a denial | A person, with AI drafting from the denial reason |
Generating the paperwork is already real. The first version of RadiusDocs, the platform Cream Digital built for Radius TeleMed, generated prior authorizations, medical necessity letters, and other documents from uploaded records. Reading records at volume is routine too: see AI document review inside a personal injury firm.
What changes for prior authorization in 2026 and 2027?
Three changes reshape the work:
- Faster payer decisions, since January 1, 2026. Under the CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F), Medicare Advantage plans, Medicaid and CHIP programs and their managed care plans, and plans on the federal exchanges must give a specific reason for every denial. All but the exchange plans must also decide expedited requests within 72 hours and standard requests within 7 calendar days. Payers post their prior authorization metrics publicly each year. Drugs are not included.
- Prior authorization APIs, starting January 1, 2027. The same rule requires those payers to offer a Prior Authorization API, so software can check requirements, submit, and track requests directly instead of through portals and fax. For managed care and exchange plans, it applies to plan years starting on or after that date. CMS has announced no delay.
- Insurer pledges. In June 2025, health plans covering about 257 million people pledged through AHIP to reduce the scope of prior authorization, honor existing authorizations for 90 days when a patient switches plans, and answer at least 80% of electronic approvals that arrive with complete documentation in real time in 2027. In April 2026, AHIP reported 11% fewer prior authorizations, about 6.5 million fewer.
Traditional Medicare is moving the other way in places. The CMS WISeR model, running from January 2026 through 2031 in New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington, uses technology companies with AI, plus human clinical review, for prior authorization of selected services. Licensed clinicians make every denial decision.
What should a practice do now?
Get the workflow ready for the APIs before they arrive:
- Document how prior authorization works today, payer by payer, using the method in how to document a workflow for AI.
- Track every request and its outcome, including denials by reason. Patterns show which payers and services to automate first.
- Start with tracking and preparation, which carry the least risk and save the most time.
- Keep submission in the approval lane until the requests come back approved, week after week.
Eligibility is usually the step before all of this, covered in how automated insurance eligibility verification works. That split, AI on preparation and tracking and trained people on approvals and follow-up, is what Cream Digital's Back Office Operations is built around.
Key facts
- Practices complete an average of 40 prior authorizations per physician per week, and 26% of physicians report one led to a serious adverse event.Source: AMA prior authorization physician survey, 2025
- Since January 1, 2026, covered payers must give a specific reason for every prior authorization denial, and most must decide expedited requests in 72 hours and standard ones in 7 days.Source: CMS-0057-F
- Covered payers must offer a Prior Authorization API starting January 1, 2027.Source: CMS-0057-F
- Health plans pledged to answer at least 80% of electronic prior authorization approvals with complete documentation in real time in 2027.Source: AHIP, June 23, 2025
Frequently asked questions
Can AI submit prior authorizations on its own?
It can prepare them completely, but in most practices a person should review and submit. A wrong submission means a denial, a delay for the patient, and rework, so prior authorization belongs in the approve-before-sending lane.
Will payer APIs remove the need for automation?
They make it easier. Once payers offer standard prior authorization APIs, software can check requirements and submit requests directly instead of through portals and fax, which is exactly the kind of connection an automated workflow uses.
What is the fastest win for a practice?
Tracking. Many authorizations stall because nobody follows up. An automated status check with alerts for pending, more-information, and denied requests catches problems days earlier than a weekly manual review.