The American Medical Association says one physician and their staff handle 40 prior authorizations in a typical week. The work consumes 13 hours. On July 31, CMS finalized another piece of the electronic prior authorization system that healthcare organizations will soon be expected to use.
Thirteen hours is nearly two working days spent checking rules, assembling records, submitting requests, and chasing status updates. Multiply it across ten physicians and the practice is carrying 130 hours of weekly administrative work.
We think many healthcare teams will make the wrong move here. They will buy an AI tool that promises to automate the whole decision. The better first investment is less dramatic: connect the systems, remove repeat data entry, and keep clinical judgement visible.
What changed on July 31
The Centers for Medicare and Medicaid Services published its fiscal year 2027 hospital payment final rule on July 31. The Office of the National Coordinator for Health Information Technology, known as ONC, used that rule to adopt updated standards for electronic prior authorization and related data exchange.
These standards describe how healthcare software should exchange information through APIs. An API is simply a controlled way for one system to request data or an action from another system.
The technical foundation is FHIR, pronounced "fire." It is a common format for exchanging healthcare data. Instead of every provider, payer, and software vendor inventing a separate connection, FHIR gives them a shared structure for information such as a patient, treatment request, coverage rule, or authorization response.
ONC's adopted standards support three practical steps:
-
Discover whether a service needs prior authorization and what the payer requires
-
Collect the relevant documentation from the clinical record
-
Submit the request and receive a structured response
CMS says electronic prior authorization interfaces from impacted payers go live on January 1, 2027. The newly finalized hospital measure is optional for calendar year 2027 and mandatory beginning in 2028.
That sequence matters. 2027 is not a distant compliance date. It is the year healthcare teams should prove that the workflow works before reporting becomes mandatory.
The business case is bigger than faster submission
The obvious saving is staff time. The AMA's 2025 survey found an average of 40 requests and 13 physician and staff hours per week. Forty percent of physicians reported employing staff who work exclusively on prior authorization.
Electronic exchange will not erase all 13 hours. Some requests genuinely need clinical review, missing records still need investigation, and payer rules will not suddenly become simple.
But the repetitive part is a good target.
Suppose a ten-physician group spends 130 hours a week on prior authorization. If better system connections remove only 30% of that workload, the practice recovers 39 hours each week. At a blended staff cost of $35 per hour, that is roughly $71,000 of annual capacity across 52 weeks.
That is not automatically $71,000 in cash savings. The practice may use the time to answer patients faster, reduce overtime, or process more appointments with the same team. We prefer that honest framing. Capacity is valuable, but it is not the same as laying off a full-time employee.
There is a revenue angle too. A request stuck because one attachment is missing can delay a procedure or cause the patient to abandon care. Faster completeness checks and clearer status tracking reduce the time between a clinical decision and scheduled treatment.
CMS estimates its broader electronic prior authorization policies will save about $15 billion over ten years. A single provider should not copy that figure into a business case. The useful number is local: minutes spent per request, rework rate, denial rate caused by missing information, and days from order to decision.
Do not start with AI deciding what gets approved
Prior authorization is often presented as an obvious AI use case. That description is too loose.
AI can read an incoming document, classify the service, find a missing field, summarize a clinical note, or suggest which record supports a request. Those tasks help a person prepare and review the case.
An autonomous system making adverse clinical or coverage decisions is a different proposition. It carries patient-safety risk, requires stronger evidence, and can make a bad process fail faster.
Six in ten physicians in the AMA survey said they were concerned that AI could increase denial rates. That concern is reasonable. If a model learns from inconsistent past decisions, automation can reproduce the inconsistency at scale.
We would not begin by automating denials. We would begin with the work that both sides should agree is wasteful:
-
Checking whether authorization is required
-
Pulling known fields from the electronic health record
-
Flagging missing documents before submission
-
Tracking status without phone calls or portal checks
-
Routing exceptions to a named employee
The system can prepare. A qualified person should remain accountable for clinical judgement and adverse decisions.
The workflow matters more than the model
A prior authorization project is not successful because a model can summarize a PDF. It succeeds when the request moves from the clinician's order to the payer and back without staff copying the same facts across disconnected screens.
Map one request from start to finish. Record every system involved, every manual handoff, and every point where staff wait for information. The map will usually include an electronic health record, a payer endpoint, document storage, scheduling, and a work queue.
Then design the smallest complete path.
-
A clinician places an order.
-
The system checks the payer's coverage requirements.
-
Required information is assembled from approved sources.
-
A staff member reviews the packet and submits it.
-
The response returns to the same work queue.
-
The scheduling team and patient receive the appropriate update.
Do not call a project complete if step four is electronic but staff still phone the payer for status and manually update scheduling. That is a digital form, not an automated workflow.
Measure a baseline before buying software
For four weeks, track a small set of numbers for one specialty or service line:
-
Requests per week
-
Staff minutes per request
-
Requests returned for missing information
-
Time from order to payer decision
-
Number of manual status checks
-
Appeals and overturned denials
Those numbers create a defensible budget. They also reveal whether the first problem is software, payer variation, or a messy internal process.
A practice sending 30 requests a month may be better served by its existing EHR vendor's certified module. A health system processing thousands of requests across several payers may justify a custom integration and a shared operations layer.
Buying a standard connection is usually smarter than rebuilding FHIR plumbing from scratch. Custom development is useful around the connection: routing work across departments, joining data from older systems, enforcing review rules, and showing leaders where requests stall.
What electronic prior authorization will not fix
Standard APIs do not make payer policies consistent. They do not guarantee approval, and they cannot turn incomplete clinical documentation into good evidence.
They also do not remove security obligations. Prior authorization data contains protected health information. Limit access, keep an audit log, and test the controls before launch. A convenience integration that exposes more patient data than necessary is not progress.
Implementation will vary by payer, EHR, and program. Healthcare organizations should confirm the exact CMS requirements that apply to them and involve compliance counsel where interpretation is needed. Software teams can implement the workflow. They should not pretend to replace legal or clinical judgement.
The practical opportunity is narrower and still valuable. Remove duplicate entry. Catch missing information before submission. Return status to the team that needs it. Keep a person responsible when the case is not routine.
Axentia builds AI applications and full-stack integrations for operational workflows where data, approvals, and human decisions have to work together. If your team is preparing for electronic prior authorization and wants to prove one service line before a larger rollout, book a call with us. We can map the current process and identify the first hours worth recovering.