The short answer
The most-cited prior authorization burden figures come from the American Medical Association’s physician surveys. In its survey of 1,000 physicians in late 2024, practices completed 39 prior authorization requests per physician per week, physicians and staff spent 13 hours a week on them, and 40% of physicians had staff who work only on prior authorizations.1 In the same survey, 61% of physicians said they’re concerned that health plans’ use of AI is increasing denials.2 From 1 January 2026, CMS-0057-F requires impacted payers to decide urgent requests within 72 hours and standard ones within seven days, and from 2027 to offer a FHIR prior authorization API.3
Deedglass
A denial needs two different people
Drafted by AI
Prior authorization
MRI, lower back · draft: deny
Request 7781 · made-up member
A denial needs two different people
Approvals · 1 of 2
- RIR. IyerNurse reviewerApproved
- AMDr. A. MehtaMedical directorTo decide
Runs only after both approve
- One person can’t approve twice
- One refusal is final
- An agent can never approve itself
The numbers
| Measure | Figure | Source |
|---|---|---|
| Prior authorizations per physician per week | 39 | AMA survey, late 20241 |
| Physician and staff hours per week | 13 | AMA survey, late 20241 |
| Physicians with staff working only on prior auth | 40% | AMA survey, late 20241 |
| Say prior auth delays access to necessary care | 93% | AMA survey2 |
| Say requests are often or always denied | 31% | AMA survey2 |
| Say denials increased over five years | 75% | AMA survey2 |
| Concerned payer AI increases denials | 61% | AMA survey2 |
For context on denials more broadly (all claims, not only prior authorization): a Premier survey of over 500 providers published in March 2024 found nearly 15% of claims to private payers are initially denied5, and Kodiak Solutions found an initial denial rate of 11.8% in 2024 data.6
How to read them
These are useful figures with real limits.
- They’re self-reported. Physicians estimated their own workload. Self-reports can overstate or understate; there’s no time-and-motion measurement behind them.
- The AMA is an advocacy body campaigning for prior authorization reform. The survey is still the best-known source, but cite it as the AMA’s survey, not as a neutral census.
- “Concerned” isn’t “measured”. The 61% figure is about physicians’ worry that payer AI raises denials, not a measurement that it does.2
- Figures move between surveys. AMA surveys in different years give somewhat different weekly counts and hours. Always quote the year.
- Prior authorization isn’t the same as claim denials. The Premier and Kodiak figures cover claims generally. Don’t merge them into one number.
What the time costs: a worked example
Turning hours into money depends on your own staffing costs, so here’s the method rather than a headline.
| Input | Value | Note |
|---|---|---|
| Physicians in the practice | 10 | Your number |
| Hours per physician per week | 13 | AMA survey1 |
| Working weeks per year | 48 | Assumption |
| Loaded cost per staff hour | Your figure | Use your payroll data |
Ten physicians × 13 hours × 48 weeks = 6,240 hours a year spent on prior authorization by physicians and staff together. Multiply by your loaded hourly cost. Remember the 13 hours mixes physician and staff time; if you know your split, cost them separately.
The number is useful for one decision: how much automation or process change is worth. It’s not a claim about any specific practice.
If you want a better input than the survey average, time it. Ask staff to log prior authorization work for two ordinary weeks: each request, the payer, minutes spent, and whether it needed a phone call or a fax. Two weeks of your own logs will tell you more than any national figure, and they show which payers and which services take the most time, which is where automation pays first.
What changes from 2026
CMS-0057-F applies to Medicare Advantage organisations, state Medicaid and CHIP fee-for-service programs, Medicaid managed care plans, CHIP managed care entities and, for some provisions, Qualified Health Plan issuers on the federal exchanges.3
| Date | Requirement |
|---|---|
| 1 January 2026 | 72-hour urgent and 7-day standard decisions; a specific reason for every denial3 |
| 31 March 2026 | First public prior authorization metrics3 |
| 1 January 2027 | FHIR prior authorization API3 |
The rule’s policies don’t cover prescription drugs.3 The API is built on HL7’s Da Vinci guides; the submission guide, PAS, returns either a decision or a pended status that’s resolved later.4 Our Da Vinci guide post explains how CRD, DTR and PAS fit together.
The published metrics will give the first payer-reported figures to set beside physician surveys. When they appear, compare like with like: payer decision times measure something different from practice staff hours.
Where automation fits, and the AI worry
Automation can cut the 13 hours on the provider side: discovering requirements in the EHR, pre-filling documentation from the record for a person to confirm, submitting electronically, and tracking pended requests. That’s what the Da Vinci guides are designed for.
The AI worry is about the other side: payers using automated tools to deny. The rule’s response is structural, not a ban: every denial needs a specific reason.3 For anyone building automation on either side, that implies records. Keep the criterion behind each decision, who reviewed it, and what the system actually sent and received, so a denial can be traced back to a rule and a person. Our CMS-0057-F post covers the denial-reason design in detail.
Questions people ask
How many prior authorizations do physicians do per week?
39 per physician per week, in the AMA’s survey of 1,000 physicians in late 2024.1
How much time does prior authorization take?
13 hours of physician and staff time per week, in the same survey.1
Does AI increase prior authorization denials?
The AMA survey found 61% of physicians are concerned it does.2 That’s a measure of concern, not of effect.
Payers and vendors can see how automated actions can be recorded and checked on our insurer page, with the record format in the provenance spec. To try approvals and read-backs on a real database in your browser, open the playground; to grade what an agent can reach on your own machine, run npx deedglass scan (the grade, explained).
Written with AI assistance and checked against our claims register.