Payer AI Systems Research
Systematic research into how health plans build and operate AI systems for utilization management — specifically auto-authorization and appeals workflows. Directly informs our Premera engagement and positions us to replicate across payers.
Why This Exists
Premera told us (Feb 11, 2026):
- They have auto-auth live for ~6 conditions, scaling to hundreds
- Appeals going into production March 1st — providers are 1 year ahead with AI-generated appeals
- They use Elementum for workflow automation, Facets (IBM) for claims, Phoenix for AI tracing
- Data was "built for a human to do a query, not for an agent to go find it"
We need to show up day 1 at Premera understanding the landscape, not learning it on their dime.
Research Tracks
| Track | Directory | Focus |
|---|---|---|
| Payer Tech Stacks | stacks/ | Core admin systems (Facets, QNXT, TriZetto), how data flows, common architectures |
| Data & Permissioning | data-access/ | VPC constraints, AI security gateways, PHI handling in LLM pipelines, HIPAA for GenAI |
| Auto-Authorization | auto-auth/ | How payers scale auto-auth, criteria engines (InterQual/MCG), medical policy digitization |
| Appeals AI | appeals/ | Provider-side AI tools, payer response systems, workflow automation (Elementum etc.) |
| Cross-Payer Patterns | cross-payer/ | How BCBS plans, UHC, Cigna, Aetna approach UM automation |
How to Use
Run the agent prompt at docs/agent-prompts/payer-ai-research.md against any track. Each track produces findings in its directory. Synthesis goes in synthesis.md at this level.
Key Questions We're Answering
- What does a typical payer UM tech stack look like end-to-end?
- How do you get AI into a payer's VPC and through their security gateway?
- What's the state of the art for auto-auth at scale (beyond 6 conditions)?
- What AI tools are providers using to generate appeals, and what are payers building to counter?
- What patterns work across multiple payers vs. what's Premera-specific?