How a regional health system cut its prior authorization cycle time by 80% — from 11 days to 2.1 — and freed clinical staff from one of the most burdensome administrative workflows in U.S. healthcare.
Get StartedPrior authorization is one of the most universally disliked processes in American healthcare. Physicians dislike it because it delays care. Clinical staff dislike it because it eats hours of repetitive, rule-following work. Patients experience it as an unexplained delay between a recommended treatment and receiving it.
Using an AI agent architecture, this health system reduced its median prior authorization cycle time from 11 days to 2.1 days, while cutting the clinical staff time spent on authorization tasks by 65%.
The health system processed ~4,500 authorization requests per month across its specialty practices, with a median 11-day cycle time from submission to payer decision. Roughly 30% of requests were returned for additional documentation, adding ~4 days each as staff regathered records and resubmitted.
A team of 14 specialists spent most of their day in payer portals, gathering EHR documentation and writing justification summaries. Payer-specific expertise lived in individual heads, not documented systems — so when experienced staff left, their knowledge left with them.
Delays were financially measurable too: authorization processing added ~8 days to order-to-procedure time for the ~60% of orders requiring it. For high-value imaging, surgery, and specialty medications, each day of delay meant rescheduled procedures, patients seeking care elsewhere, and missed clinical windows.
The agent is only as accurate as the criteria it matches against. Ambiguous, outdated, or incomplete criteria raised payer return rates — keeping the library a living resource demanded ongoing investment the budget had underestimated.
The specialists knew the payer quirks and edge cases the AI handled poorly. Their pilot-phase feedback drove dozens of refinements — automation built without the people whose work it changes ends up functional but operationally suboptimal.
First-pass approval rose from 71% to 79% because AI-prepared justifications were more complete and consistent than manual ones. Faster cycle time plus higher approval produced a compounded benefit bigger than either metric alone.
Payers with FHIR-based prior auth APIs (per the CMS Interoperability Final Rule) enabled the fastest electronic submission and tracking. For portal-only payers, automation still helped — just not as dramatically.
Routine cases were processed end to end by the agent; exceptions were routed to specialists with full context. That split cut average specialist time per authorization from 38 minutes to 13.
On request, the agent pulled the exact clinical documentation each payer requires for that procedure straight from the EHR — eliminating ~25 minutes of manual compilation per request.
It compared the record against payer coverage criteria and drafted a justification letter citing the specific findings that met them — flagging borderline or ambiguous cases for closer human review.
FHIR-capable payers received electronic submissions with automatic status checks written back to the EHR; portal-only payers got pre-populated fields, reducing data entry to a review-and-submit step.
Borderline cases, documentation gaps, and payer info requests were routed to specialists with the AI's assessment, the issue, and the relevant criteria attached — resolving exceptions without starting from scratch.
