Healthcare's back office runs on a paradox: the industry spends roughly a trillion dollars a year on administration, a quarter to a third of all U.S. health spending, and yet the defining experience of that administration, for the people inside it, is manual work. Nowhere is that clearer than prior authorization.
The American Medical Association's 2025 physician survey puts numbers on it: practices complete about 39 prior auth requests per physician per week, consuming 13 hours of physician and staff time, and 40% of physicians employ staff who work exclusively on prior auth. When some sixty insurers pledged in June 2025 to streamline the process, only a third of physicians believed it would meaningfully help.
What those 13 hours are made of
Strip away the policy debate and prior auth is a screen workflow, one of the most repetitive in any industry:
- Check whether the payer requires auth for this code, on this plan, this year
- Verify eligibility and benefits in the payer portal or clearinghouse
- Assemble the clinical documentation the payer's criteria ask for
- Key the request into the payer's portal (each payer's portal is different)
- Check status. Check status again. Answer the request for more information
- Log the outcome in the EHR and tee up the appeal if it's denied
McKinsey estimates 50–75% of the manual steps in prior auth are automatable with current AI, and notes that provider adoption of fully electronic prior auth is still under half, with about a third of requests handled entirely manually.
The stakes of getting it wrong
The cost of manual process isn't only the labor going in. It's the rework coming back. Hospital claim denial rates hit 11.6% in 2025, and 41% of providers now run denial rates above 10%, up from 30% in 2022. Reworking a single denied claim costs $25 to $181 depending on complexity. Much of that traces back to front-end administrative misses: an eligibility detail, a missing auth, a code that changed.
The industry-wide ledger, tracked annually by CAQH, shows what automation is worth here: $258 billion in administrative costs avoided in the 2025 index, with eligibility verification the single largest remaining savings pool. A manual eligibility check costs a provider $7.97 per transaction against $2.18 fully electronic.
The bottleneck isn't knowledge; everyone knows the payer's portal and the criteria. The bottleneck is that a skilled human has to sit at the screen and walk through them, forty times a physician, every week.
Where agents fit, carefully
Healthcare is exactly where "human in the loop" has to be a design principle rather than a slogan. Clinical judgment isn't the automation target; the portal choreography around it is. An agent taught from a recording of your own prior-auth specialist can run the checkable parts (eligibility lookups, status checks, form-filling from documentation a human assembled) and hand back a queue where every item is either done or annotated with exactly what a person needs to decide.
Health systems are already moving this direction: 80% were piloting or implementing generative AI in revenue cycle work in 2025, up from 58% two years earlier. The regulatory clock helps too: CMS rules now require payers to answer urgent requests within 72 hours starting in 2026, which only sharpens the premium on submitting clean requests fast.
The 13 hours aren't going to zero. But the share of them spent copying data between an EHR and a payer portal can, and that's the share nobody went to medical school, or nursing school, or even a weekend billing seminar, to do.
Written by Madhavam Shahi, teaching agents to run the back office at AltOps.
