Inputs & data contract
Patient/plan, ordered service, diagnosis/procedure context, provider/site, payer requirements, clinical documents, service date and existing authorization.
PRO · RCAI Agent Blueprint
Prior Authorization Agent
Determine authorization requirements, assemble supported evidence, track status and prevent avoidable authorization-related denials.
Service Planned → Requirement Check → Not Required / Required → Evidence Gathered → Submitted → Pending / More Info → Approved / Denied → Scheduled / Escalated
Tool interfaces
EHR/PMS, payer policy/rules source, portal/API adapter, document workflow, status queue and audit log.
Decision policy
Requirement logic must be grounded in current payer/plan evidence. Model reasoning can organize documentation but cannot invent clinical facts, criteria or authorization approval.
Exception matrix
Plan ambiguity; no policy evidence; urgent service; missing clinical documentation; peer-to-peer; partial approval; expiration; service/code change; payer downtime.
Human-in-the-loop gates
Clinical criteria, denial response, peer-to-peer, unsupported policy interpretation and changes affecting approved scope require human ownership.
Production KPIs
Requirement accuracy; auth-related denial rate; submission completeness; approval turnaround; expired auth rate; manual touches; services protected.
Production implementation pack
Use the state model above as the control plane. The production pack adds transition guards, source precedence, retry/idempotency rules, exception ownership, audit requirements, evaluation cases and go-live acceptance thresholds.
Evaluation & acceptance
Validate classification and next-action accuracy separately. Track unsafe autonomous actions as a zero-tolerance metric, regression-test every rule or model change, and require supported evidence for every payer-specific conclusion.
Pro implementation note: adapt the blueprint to your contracts, payer policies, systems, security model and approval thresholds before production autonomy.