AI could help health plans simplify prior authorization and meet new CMS rules
Standard prior-authorization decisions for federally regulated health plans now have a seven-day limit. Expedited requests remain at 72 hours. Miss the mark and you risk penalties. That's the operational reality as of January 1.
By March 31, 2026, plans must also publish average turnaround times, denial rates, appeals, and overturns on their websites. Expect consumers, brokers, and provider partners to use those numbers to compare plans. The final rule is also pushing more consistent electronic data exchange.
What this means for executives
Prior authorization is still a heavy lift: long clinical PDFs, varying plan criteria, and back-and-forth with providers. Adding more nurses and physicians won't scale. You need better throughput, consistent decisions, and audit-ready documentation-without eroding clinical quality.
Where AI actually helps
Generative and agentic AI can reduce the grunt work. It can extract key facts from records, map them to medical-necessity criteria, and surface a clear recommendation. Routine, low-risk cases can move to auto-approval with guardrails. Ambiguous or high-risk cases get routed to clinicians with a concise, explainable summary.
Used well, AI increases speed, consistency, and traceability. It can also help plans comply with decision timelines and public reporting. Clinical oversight remains non-negotiable.
What leaders are seeing
In a 2026 survey, 93% of health plan executives said AI will add value by automating prior authorization. Investment levels vary, and large nationals are further along. Off-the-shelf, scaled solutions across all service categories are still limited, and many plans lack the clinical inputs needed to train and validate models.
Compliance and operating levers you control
- Limit prior auth where it doesn't pay off: low-cost, low-denial, low-clinical-risk services.
- Expand gold-carding for high-performing providers to reduce friction and cycle time.
- Standardize criteria and documentation checklists to cut incomplete submissions.
- Push electronic submissions and integrated data exchange to reduce manual work.
Medicare's WISeR pilot is a signal
CMS launched a pilot using AI to support prior-authorization decisions in traditional Medicare across six states. It targets services prone to fraud, waste, and abuse (e.g., skin and tissue substitutes, nerve-stimulator implants, knee arthroscopy). Inpatient-only and emergency services are excluded. Clinicians stay in the loop, and only licensed clinicians issue denials.
Takeaway: AI can sit inside the review process with clear guardrails, service scoping, and clinical oversight. Expect broader adoption if results hold.
A pragmatic AI roadmap
Next 90 days
- Reg readiness: lock in a seven-day SLA playbook; map escalation paths for day 5-7 cases.
- Service scoping: identify categories for auto-approval, clinical review, and exclusion.
- Data foundation: centralize policies, criteria, and historical decisions; index clinical PDFs.
- Pilot AI-in-the-loop: start with RAG-based evidence extraction, criteria matching, and explainable summaries for reviewer support.
- Governance: define human-in-the-loop rules; require clinician sign-off for denials.
Next 12 months
- Expand coverage: add more services with clear criteria to AI-supported review.
- Auto-approve guardrails: allow auto-approvals where evidence clearly meets policy; keep denials clinician-led.
- Provider collaboration: share checklists and feedback to reduce incomplete submissions.
- Interoperability: advance APIs and FHIR integration to reduce manual attachments and rekeying.
- Controls: build auditable logs, bias checks, and model-monitoring tied to overturns and quality outcomes.
Metrics to run the program
- Turnaround time: average, 90th percentile, and day-7 backlog.
- Auto-approval rate for low-risk services and associated clinical exceptions.
- Denial rate, appeal rate, and overturn rate by service and provider segment.
- Clinician hours per authorization and total cost per case.
- Provider resubmission rate and call volume tied to prior auth queries.
What good looks like
- Simple services: automated intake, evidence extraction, and instant approval when criteria are met.
- Complex cases: AI-generated, evidence-linked summaries for reviewers; clear rationale for decisions.
- Transparency: decision trails, policy citations, and patient/provider-friendly explanations.
- Safety: no auto-denials; clinical oversight on edge cases; continuous validation against overturns.
Common pitfalls to avoid
- Letting vendors define your criteria logic-codify your own policies first.
- Skipping clinician training data-models need real examples and nuanced judgment.
- Overextending: start with a few services, prove accuracy, then scale.
- Ignoring provider experience: reduce document burden and clarify exactly what "complete" means.
Bottom line
The seven-day clock and public reporting raise the bar for speed and consistency. Adding staff alone won't get you there. AI can streamline evidence gathering and decision support, while policy changes (like gold-carding and service scoping) trim unnecessary work. Keep clinicians in control and build for auditability. That combination can reduce cost, improve patient access, and keep you on the right side of the rule.
Upskill your team
If you need to get leaders and clinical ops teams fluent in practical AI, explore role-based options here: AI courses by job.
Your membership also unlocks:
