Skip to main content
HEA-03 · Healthcare Providers

Prior Authorization Copilot

Outcome​

Shorten prior authorization turnaround by matching the ordered service to payer medical-necessity criteria, assembling the exact clinical evidence required, and letting a utilization specialist approve the packet before submission.

Portfolio attributeValue
Primary personaUtilization specialist
Primary aggregateAuthorizationRequest
Cosmos DB containers2
Grounding dataclinical records, payer criteria, orders, codes
Development estimateUSD 60-220 monthly dev estimate
Runtime optionsPython FastAPI and C# ASP.NET Core

Productivity hypothesis​

These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.

Impact typeMeasureConservative internal assumptionCustomer measurement
ProductivitySubmission packet preparation time7-12% reduction(baseline median packet-preparation minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskCorrected or reopened packets2-5% reduction(baseline corrected or reopened packet rate - assisted rate) / baseline rate * 100

Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.

Rationale: Clinical records, payer criteria, orders, and codes can reduce authorization-packet assembly effort.

Pilot guardrail: Utilization specialists retain payer-criteria interpretation and submission approval.

Customer validation plan​

  1. Baseline the current workflow with representative users and tasks.
  2. Run the accelerator on the same task mix with human review retained.
  3. Measure submission packet preparation time and corrected or reopened packets with the documented formulas.
  4. Replace internal assumptions with customer-observed results before a business case.

Why Azure Cosmos DB​

This AI solution needs an operational data plane, not only a vector index:

  • Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
  • Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
  • Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
  • Change feed reacts to source and operational updates without polling.
  • Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.

Domain workflow​

Domain workflow

StageActorInput → outputApproval
1. Receive the order and identify payer policyOrder intake serviceService order with diagnosis and payer plan → AuthorizationRequest bound to a payer medical policyPolicy controlled
2. Decompose the policy into testable criteriaCriteria extraction agentPayer medical policy text → PayerCriteriaClause set for the servicePolicy controlled
3. Retrieve matching clinical evidenceEvidence retrieval agentCriteria clauses and patient chart → ClinicalEvidenceItem set with excerpts and datesPolicy controlled
4. Adjudicate criteria and identify gapsAdjudication agentCriteria and retrieved evidence → CriteriaMatchResult set with missing elementsPolicy controlled
5. Approve and transmit the packetUtilization specialistAssembled packet with match results → Packet transmitted to the payer with a reference numberHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
authorizationRequests/tenantId, /authorizationIdAuthorizationRequest, ClinicalEvidenceItem, CriteriaMatchResult, SubmissionPacket/excerptVectorEmbedding, /excerptVector
payerPolicyLibrary/tenantId, /payerPolicyIdPayerCriteriaClause/criteriaVectorEmbedding, /criteriaVector

Query contract​

QueryContainerScopePurpose
Authorization packet assemblyauthorizationRequestssingle-partitionRead the request, the gathered clinical evidence, and the criteria match results for one authorization in a single partition-aligned query. CriteriaMatchResult denormalizes the clause reference and the quoted criterion text captured at match time, so the packet never joins to the payerPolicyLibrary container at assembly time.
Evidence gap worklistauthorizationRequestssingle-partitionList criteria that are unmet or ambiguous so staff can chase the missing document.
Urgent pending determinationsauthorizationRequestscross-partition-boundedBounded cross-partition control-plane query: track expedited requests awaiting a payer decision against the turnaround clock. It cannot carry the second partition key because the turnaround clock is monitored across every request in flight, so it runs as a tenant-scoped scheduled sweep off the interactive path, bounded by expedited requests inside the 72-hour statutory turnaround window, executed with MaxItemCount paging and continuation tokens.
Denial pattern analysisauthorizationRequestscross-partition-boundedBounded cross-partition control-plane query: group recent denials by payer and service code to find recurring documentation failures. It cannot carry the second partition key because pattern analysis is inherently a portfolio-level analytical question, so it runs as a tenant-scoped scheduled sweep off the interactive path, bounded by denied determinations inside a 90-day window, aggregated server side, executed with MaxItemCount paging and continuation tokens.

Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.

Azure deployment​

Azure deployment

Clone the source repository, authenticate with Azure, and run:

Set-Location accelerators/healthcare-providers/prior-authorization-copilot
az login
./Deploy.ps1 -Runtime python -Location eastus2

Change python to csharp to deploy the .NET implementation.

Explore the implementation​

Return to all accelerators