📋 Executive Overview
Southern Cross Health — Agentic AI Transformation Roadmap
Southern Cross's challenge is not simply claims processing — it is unlocking and validating information distributed across SharePoint and document repositories. By first establishing a Document Intelligence foundation, then automating validation, and finally introducing AI-assisted claim decisioning, Southern Cross can significantly reduce manual effort, improve compliance, and accelerate member claim outcomes.
— Executive Positioning Statement
Avg Claim Review Time
25min
↓ Target: 5–10 mins with AI
Document Search Time
10min
↓ Target: <1 min with AI Search
Straight-Through Processing
20%
↑ Target: 50–60% with AI
Assessor Productivity
+50%
↑ Target gain from AI automation
Current State → Future State
CURRENT STATE
SharePoint Repositories
Emails
Provider Documents
Medical Reports
Invoices & Referrals
⬇ Manual Search & Review
⬇ Manual Validation
⬇ Claims Assessment
⬇ Member Outcome
→
PROJECT 1
Document Intelligence Platform
Document Discovery Agent
Document Classification Agent
Document Summarisation Agent
Knowledge Retrieval Agent
→
PROJECT 2
Document Validation Agent
Policy Validation Agent
Evidence Validation Agent
Provider Validation Agent
Compliance Agent
Fraud Detection Agent
→
PROJECT 3
Claims Recommendation Agent
Claim Assessment Agent
Claim Recommendation Agent
Decision Explanation Agent
Member Communication Agent
Payment Trigger Agent
→
OUTCOME
Member Notification
Payment Processing
Audit Trail
Compliance Record
Project 1 — Find the Information
Document Intelligence Platform
Eliminate time spent searching SharePoint and reading large document packs. AI retrieves, classifies, and summarises all relevant claim documents in under 60 seconds.
Project 2 — Validate the Information
Document Validation Platform
Reduce manual checks and improve compliance and accuracy. Automated validation of policy rules, evidence completeness, provider eligibility, and fraud indicators.
Project 3 — Decide Using the Information
AI-Assisted Claims Decisioning
Accelerate claims decisions and improve member experience. AI generates transparent recommendations with confidence scores, reasoning, and member notifications.
🏗️ Platform Architecture
Technology stack recommendation for the Southern Cross AI Claims prototype
Recommended 2-Week Prototype Stack
| Layer | Technology |
|---|---|
| UI / Frontend | Streamlit or React |
| Agent Framework | LangGraph / CrewAI / Semantic Kernel |
| LLM | GPT-4o (recommended primary) |
| Vector Store | Azure AI Search / Pinecone |
| Document Storage | SharePoint Online |
| OCR | Azure Document Intelligence |
| Workflow | LangGraph |
| Rules Engine | Python Rules / Drools |
| Database | PostgreSQL |
| Monitoring | LangSmith |
Minimal 4-Agent Executive Demo
If you need a quick executive demo, build only these 4 agents for maximum impact:
| Agent | Complexity | Demo Value |
|---|---|---|
| Document Discovery Agent | Medium | High |
| Document Summarisation Agent | Low | Very High |
| Policy Validation Agent | Medium | Very High |
| Claim Recommendation Agent | Medium | Very High |
Example System Prompt — Claim Recommendation Agent
# ─────────────────────────────────────────────────────────────────
# SYSTEM PROMPT: Southern Cross Health — Claims Assessment Agent
# ─────────────────────────────────────────────────────────────────
role: You are a Southern Cross Health Claims Assessment Agent.
review:
1. Claim summary and supporting documents
2. Policy coverage, product type, and effective dates
3. Waiting periods for the claimed procedure
4. Benefit limits and sub-limits
5. Provider eligibility and network participation
6. Missing or incomplete documentation
output_format:
Decision: Approve | Partially Approve | Reject | Escalate
Reasoning:
- Coverage assessment
- Policy validation findings
- Missing evidence observations
- Risk observations
Confidence Score: 0–100%
Recommended Reimbursement Amount: $NZD
Next Action Required:
# Always ground your decision in Southern Cross policy documentation.
# Do not approve claims that fall within exclusion periods.
# Flag any provider not registered in the provider master data.
🤖 Agent Catalogue
Full agentic AI capability map across all three projects
| Project | Business Capability | Agentic AI Agent | LLM Role | Enterprise Data Sources | AI / Technology | Output |
|---|---|---|---|---|---|---|
| Project 1 Document Intelligence |
Document Intelligence Platform | 🔍 Document Discovery Agent |
Understand user intent, semantic search, contextual retrieval | SharePoint, Email, Claims Repository, Provider Docs, Policy Docs | GPT-4o Claude Sonnet Azure AI Search RAG | Relevant document list |
🏷️ Document Classification Agent |
Identify document types and business context | PDFs, Invoices, Referrals, Medical Reports | GPT-4o Mini Azure Document Intelligence | Classified document catalogue | ||
📝 Document Summarisation Agent |
Generate consolidated claim summaries | Retrieved documents | GPT-4o Claude Sonnet | Single claim summary | ||
💬 Knowledge Retrieval Agent |
Answer questions over enterprise documents | SharePoint Knowledge Base | RAG LLM | Natural language responses | ||
| Project 2 Validation |
Document Validation Platform | 📋 Policy Validation Agent |
Compare claim information against policy rules | Policy schedules, benefits, exclusions | LLM Rules Engine | Coverage validation report |
🔎 Evidence Validation Agent |
Verify mandatory evidence exists | Claims documents | LLM Workflow Engine | Missing evidence report | ||
🏥 Provider Validation Agent |
Verify provider eligibility and network participation | Provider Master Data | LLM API Integration | Provider validation status | ||
⚖️ Compliance Agent |
Check regulatory and business compliance | Internal policies and governance rules | LLM Compliance Rules | Compliance assessment | ||
🚨 Fraud Detection Agent |
Identify anomalies and suspicious claims | Historical claims data | ML Model LLM reasoning | Fraud risk indicators | ||
| Project 3 Decisioning |
AI-Assisted Claims Decisioning | 🧠 Claim Assessment Agent |
Review all validated information holistically | Validation outputs from Project 2 | GPT-4o Claude Opus 4 | Holistic claim assessment |
✅ Claim Recommendation Agent |
Recommend Approve / Reject / Escalate | Assessment results | LLM Business Rules | Recommendation + confidence score | ||
📖 Decision Explanation Agent |
Explain why a decision was made | Assessment evidence | LLM | Transparent audit trail | ||
📧 Member Communication Agent |
Draft member notifications | Decision output | GPT-4o Mini | Email / SMS / Letter | ||
💳 Payment Trigger Agent |
Initiate payment workflow | Approved claims | Workflow Platform | Payment request |
⚡ Live Claims Intelligence Demo
Enter a Claim ID to run the full 8-agent AI pipeline with synthetic data
Southern Cross — AI Claims Processor
Agentic AI pipeline · GPT-4o + Claude Sonnet 4 · Azure AI Search
LIVE DEMO
Agent Pipeline Status
Document Discovery Agent
Waiting — Search SharePoint for claim documents
WAITING
Document Classification Agent
Waiting — Classify document types
WAITING
Document Summarisation Agent
Waiting — Generate consolidated summary
WAITING
Policy Validation Agent
Waiting — Validate coverage and benefit limits
WAITING
Evidence & Missing Documents Agent
Waiting — Check evidence completeness
WAITING
Provider Verification Agent
Waiting — Verify provider registration
WAITING
Fraud Detection Agent
Waiting — Anomaly detection
WAITING
Claim Recommendation Agent
Waiting — Generate final recommendation
WAITING
Member Communication Agent
Waiting — Draft member notification
WAITING
📊 AI Assessment Results
📂 Claim Information
📄 Documents Retrieved (SharePoint)
📝 AI Claim Summary
📋 Policy Validation Results
🔎 Evidence Completeness Check
🏥 Provider & Fraud Validation
🎯 AI Confidence Score
—
Confidence
💰 Reimbursement Summary
📧 AI-Generated Member Notification
Email · Auto-generated by Member Communication Agent
📈 Business KPIs to Demonstrate
Measurable outcomes from the AI Claims Intelligence Platform
Claim Review Time
20–30 mins
⬇
5–10 mins
Document Search
10 mins
⬇
<1 min
Validation Effort
15 mins
⬇
2 mins
Assessor Productivity
Baseline
⬆
+50%
Straight-Through Processing
20%
⬆
50–60%
Processing Time Reduction
Claim Review (Current: 25 min)25 min
Claim Review (AI Target: 7 min)7 min
Document Search (Current: 10 min)10 min
Document Search (AI Target: 45s)<1 min
Straight-Through Processing Rate
50%
Target STP Rate with AI (vs 20% current)
0%Current: 20%Target: 50–60%100%
🧠 LLM Strategy
Recommended model selection by agent category
Search & Retrieval
GPT-4o
Document Discovery Agent, Knowledge Retrieval Agent. Best for understanding nuanced search intent and contextual retrieval over large document sets.
Summarisation
GPT-4o
Document Summarisation Agent. Excellent at producing concise, accurate summaries from complex medical and insurance documents.
Validation & Reasoning
GPT-4o or Claude Sonnet 4
Policy Validation Agent, Evidence Validation Agent. Both models excel at structured reasoning against rules and checklists.
Complex Assessment
Claude Opus 4 or GPT-4o
Claim Assessment Agent, Claim Recommendation Agent. Highest capability models for multi-factor decision making with confidence scoring.
Member Communications
GPT-4o Mini
Member Communication Agent, Document Classification Agent. Fast, cost-efficient for template-driven generation and classification tasks.
Knowledge Assistant
GPT-4o
Knowledge Retrieval Agent. Used for RAG-backed Q&A over SharePoint knowledge bases, policy documents, and procedure guides.
Full LLM Model Mapping
| Agent | Primary Model | Fallback Model | Rationale |
|---|---|---|---|
| Document Discovery Agent | GPT-4o | Claude Sonnet 4 | Semantic search + intent understanding |
| Document Classification Agent | GPT-4o Mini | GPT-4o | Fast, cost-efficient classification |
| Document Summarisation Agent | GPT-4o | Claude Sonnet 4 | High-quality medical summarisation |
| Knowledge Retrieval Agent | GPT-4o | — | RAG + conversational Q&A |
| Policy Validation Agent | GPT-4o | Claude Sonnet 4 | Structured policy rule checking |
| Evidence Validation Agent | GPT-4o | — | Checklist-based completeness validation |
| Provider Validation Agent | GPT-4o Mini | — | API lookup + structured verification |
| Compliance Agent | GPT-4o | — | Regulatory rule assessment |
| Fraud Detection Agent | ML + GPT-4o | — | Anomaly detection + LLM reasoning |
| Claim Assessment Agent | Claude Opus 4 | GPT-4o | Complex multi-factor reasoning |
| Claim Recommendation Agent | GPT-4o | Claude Opus 4 | Final recommendation with rationale |
| Decision Explanation Agent | GPT-4o | — | Transparent, plain-language explanation |
| Member Communication Agent | GPT-4o Mini | — | Cost-efficient, empathetic drafting |
| Payment Trigger Agent | No LLM required — workflow automation platform (Power Automate / n8n) | ||