Premera — Scope of Work
Last updated: 2025-02-10 Source: Near-final proposal draft (not yet sent/confirmed with Premera, but expected to land close to this)
Summary
We partner with Premera to reduce manual utilization review time by generating structured medical chart summaries, extracting UM-relevant flags, automating key sub-workstreams, and providing usage analytics. We integrate with Premera's data sources (EMR, secure file share, PDF packets, HL7 feeds), align outputs with their nursing workflow templates, and deliver a production-ready HIPAA-aligned application with customized development for integration and template work.
Objectives
- Reduce nurse review time → allocate towards higher-value activities
- Surface UM-specific flags (start of service timing, critical diagnoses, queue priorities) → proactive management of high-risk cases
- Structured summaries mapped to Premera's templates → cross-team consistency and quality
- Low-friction data ingestion from hospital systems / Premera intake channels → better data integration, improved hospital communication
- Analytics on usage, throughput, cost → C-level visibility, management-level ops control
In Scope
- Clinical Chart Summaries. Structured summaries from PDFs and other clinical artifacts
- Data Ingestion. Connect to agreed sources (Premera-hosted data mgmt, PDF packets, secure file share, HL7 feeds — ADT, ORU, CCDA)
- Template Alignment. Map outputs to Premera's existing summary templates and QA expectations
- Analytics. Dashboard: volume, processing time, usage cost
- Integration. Training, support, iterative tuning with nurses and executive team
Out of Scope
- Formal integration of proprietary criteria engines (MCG/InterQual)
- Automated clinical determinations without human review
Deliverables
- Discovery Summary: Workflow map, data sources, feature requests
- UM Flag Specification: Written list of UM flags + extraction logic
- Technical Brief: Agent architecture, skill orchestration, technical logic
- Summary Template Implementation: Output matched to Premera's template
- Integration: Live data ingestion from approved sources
- Dashboard: Usage + throughput analytics
- Training: 1-2 sessions for reviewers and administrators
Responsibilities
Daisy AI
- AI-driven workflow for drafting/supporting clinical reviews (human-in-the-loop)
- Summary and UM flag extraction based on Premera input
- Build and maintain ingestion pipeline(s)
- Analytics and admin visibility into usage/costs
- Onboarding, training, success review
Premera
- Nurse SME availability — for tuning, discovery, UM flag definition
- Review templates — existing summary template and QA checklist
- Hospital coordination — data source access (EMR integrations, file sharing)
- Feedback loop — validate outputs, provide iterative feedback, participate in recurring meetings
Data Access and Security
- HIPAA-aligned data processing
- Secure data transfer method approved by Premera
- Daisy AI provides security deck and standard controls for review
- BAA with Premera with detailed provisions on acceptable data use and access
Timeline (Draft)
| Phase | Weeks | Activities |
|---|---|---|
| 0: Contracting & Security | 0-4 | Exchange security materials, finalize SOW, confirm data access |
| 1: Discovery & Setup | 4-8 | Nurse interviews, UM flag definitions, template mapping, test data |
| 2: Integration & Launch | 8-12 | Data ingestion live, initial review runs, QA feedback loop |
| 3: Evaluation & Scale | 12+ | KPI review, time savings, quality, roadmap for scale |
Exact dates TBD after onboarding and access approvals.
Success Criteria
| Metric | Target | Measurement |
|---|---|---|
| Time savings | >50% reduction in summary prep | Baseline vs. tracking (20 min → 10 min) |
| Template fit | Usable with <2 min edits | Nurse feedback survey |
| UM flag accuracy | >85% reviewer agreement | Spot check sample |
| Nurse satisfaction | >8/10 post-training | Survey |
| Security | Pass review | Architecture walkthrough + questionnaire |
Discussion Topics (for working session)
Discovery: End-to-end UM workflow, current tech stack, ADT feed architecture, past deployment failures Technical: Where to start among ~12 initiatives, data requirements, ownership preferences, scope/responsibility split Business: Regulatory/compliance gates, AI governance, timeline expectations, success definition, commercial terms
Outstanding Questions for Premera
UM Workflow: Org structure, key contacts, KPIs, teams, criteria sets, vendors, tech systems, lingo, case routing, bottlenecks (initial reviews vs. P2P vs. appeals) Tech Stack: Dev standards, platform vendors, approval process, build-in-environment preference, languages/frameworks, AI vendor relationships, existing AI infra, EMR integrations Project Management: POC, update communication, milestone cadence Commercial: Contract formality needed, diligence requirements, mutual assurances Forward Looking: Broader UM transformation alignment
Our Tech Stack (as presented)
Core: Next.js 15, React 19, TypeScript, PostgreSQL (Neon), Drizzle ORM AI: Claude Sonnet 4.5 / Opus 4.5 (primary), GPT-5.2 / GPT-4o (secondary), Vercel AI SDK, specialized agents with Zod-validated schemas Security: Clerk (mandatory MFA), HIPAA-aligned architecture (row-level security, audit logging, PHI isolation), Neon branch separation (prod PHI never touches dev) Infrastructure: Vercel (hosting, edge, preview deploys), Stripe (billing) Data Ingestion: HL7 v2 parser (ADT/ORU), CCDA/CDA processing, PDF/DOCX extraction UI: Tailwind CSS v4, shadcn/ui, Radix UI, MUI DataGrid, resizable multi-pane layout