How a regional Blue plan dropped P50 PA turnaround from 11 days to 26 seconds.
Twelve weeks. Seven CPT categories. Two parallel reviewer panels. A $4.3M annual UM operation re-architected around evidence-at-decision-time rather than evidence-at-appeal-time.
What was broken, in dollar terms.
Walked into 2026 with three problems converging at the same Q4. CMS-0057-F public reporting goes live January 1. The plan was running PA on a 2012-vintage rules engine and a Lotus-Notes-shaped intake portal. Internal counsel had reviewed the metrics the plan would have to publish and concluded that the published P50 turnaround time, denial rate, and appeal-overturn rate would each be in the worst quartile nationally.
- $4.3M annual UM operating spend, 60% labor
- P50 turnaround of 11.2 days; P95 of 28 days
- Appeal-overturn rate of 71% — within 10 points of the KFF Medicare-Advantage benchmark
- Two NCQA UM survey findings open from the 2024 audit cycle
The integration, the workflow change, the 90-day arc.
Authora deployed as a delegated UM engine sitting in front of the legacy stack. The legacy system remained the system of record for membership and claim adjudication; Authora owned the PA workflow end to end.
- Full Da Vinci PAS + CRD + DTR conformance, with X12 278 bridge for the handful of legacy provider EHRs
- Medical-policy library ingested from the plan's existing criteria-management vendor; criteria adjudication runs against the plan's own policies, not a generic set
- Parallel reviewer console for the plan's existing UM nurses — every Authora decision visible side-by-side with a human review for the first 45 days
- CMS-0057-F public-reporting fields populated automatically and exported quarterly for the plan's regulatory affairs team
The seven seeded categories, by the numbers.
Pilot ran across the seven CPT categories that, in combination, accounted for roughly 62% of the customer’s PA volume. Auto-approval rates reflect the share of cases that cleared without human review under the customer’s own medical-policy library.
| CPT | Category | Auto-approval | P50 latency | Weekly volume |
|---|---|---|---|---|
| 72148 | MRI lumbar spine | 78% | 22s | 412/wk |
| 27447 | Knee arthroplasty | 61% | 34s | 118/wk |
| 78815 | PET/CT skull-to-thigh | 69% | 28s | 84/wk |
| 96413 | IV chemotherapy admin | 84% | 19s | 260/wk |
| 90867 | TMS therapy | 57% | 41s | 62/wk |
| J9271 | Pembrolizumab inj. | 82% | 21s | 146/wk |
| E2102 | DME — CGM sensor | 88% | 17s | 338/wk |
“We did not buy Authora to make denials faster. We bought it because the denials we were issuing did not survive appeal, and the regulator was about to start publishing that fact. The platform fixed the upstream evidence problem. The downstream metrics followed.”
Year-one expansion plans.
- Expand from 7 pilot categories to the plan's full PA-required list (94 categories) by Q1 2027
- Roll the delegated UM pipeline to the plan's Medicaid managed-care line
- Migrate the existing rules engine off the legacy stack; sunset the 2012 system in Q3 2027
- Co-publish the de-identified outcome data with the plan's actuarial team to the Health Affairs research desk
The trust posture is the same as the pilot.
Every pilot runs under BAA, in us-east-2, with the same SOC 2 Type II + HITRUST r2 posture documented in /trust. The pilot economics are documented in /pilot.