Genflare · Advisory

PA Workflow Analysis. Where to start.

A deep look at how Lumara's Prior Authorization team works today — and a phased plan that starts with end-to-end intake automation as the proving ground.

Scope
26 workflows · 9+ systems
Method
Scribe analysis + walkthroughs
Output
Phased automation roadmap
Audience
Lumara PA leadership
What we found

The PA team is heroic. The workflow is fighting them.

Across 26 documented workflows, a clear picture emerged: capable specialists doing high-judgment work, slowed by mechanical document handling and fractured tooling.

8+

The document assembly tax

Across nearly every workflow, a disproportionate share of time goes to downloading, naming, filing, combining, and uploading documents. The naming convention (Lastname_Firstname DOB DocType Date) is enforced manually via copy-paste in 8+ workflows.

9+

Nine systems, one team

Staff navigate Salesforce, Adobe Acrobat, Gmail, a shared file system, Availity, payer portals, the NPI Registry, and external counsel's secure file portal — sometimes inside a single workflow.

2

Two distinct automation domains

Rules-based, high-volume intake work (great for automation) versus judgment-intensive PA specialist work — clinical review, appeal strategy, payer negotiation — that requires AI augmentation, not replacement.

Where we'd start

Intake. Not the biggest opportunity. The right proving ground.

Intake is the front door of the PA process. It's also where the workflow pattern is cleanest, the risk is lowest, and the impact downstream is the most direct.

01

Rules-based and high-volume

Account creation, document download/naming/filing, Salesforce upload — clear rules, measurable output.

02

Low risk of getting it wrong

Errors are mechanical (wrong filename, wrong folder), not clinical. A safe place to test and learn.

03

Immediately measurable

Step count and time-per-task are easy to benchmark before and after. ROI shows up fast.

04

Unblocks downstream work

The Documentation Review Complete handoff is the bottleneck. Faster intake means faster PA start, which means faster patient access.

Phase 1 in detail

End-to-end intake-to-handoff. Five sub-phases.

Sequenced for incremental delivery and testability. Automates roughly 265 documented manual steps across the team's intake workflows. By the end of Phase 1, the morning's inbox arrives as PA packets in a specialist's queue.

1A

Email-to-document pipeline

Email monitoring, document download, page sorting, document classification.

1B

Filing & Salesforce write-back

Naming convention, shared drive filing, Salesforce record creation, email organization, account creation for new patients.

1C

Clinical review assistance

AI-assisted extraction of biomarker values, exacerbation history, and provider recommendation. Confidence scoring and triage recommendations.

1D

Routing & specialist handoff

Territory-to-specialist routing, PA packet assembly, automated Salesforce task creation in the right queue.

1E

Exception handling

Auto-drafted correction requests to territory managers, follow-up task creation, editable template management.

The work being automated
Creating new account & case~36 STEPS
Clinical documentation download & filing~71 STEPS
Consult upload~44 STEPS
Insurance letter filing~45 STEPS
Converting leads~12 STEPS
Email-to-Salesforce logging~8 STEPS
22 user stories with acceptance criteria, specified in the Program Design Document.
Three solution paths

Three credible ways forward. Evaluate them in parallel.

Each path has real strengths and real risks. We recommend running them side by side — fast — and letting evidence drive the decision.

A

Domain-specific PA vendor

Pre-built for healthcare PA intake
Strengths
  • Purpose-built for PA intake & extraction
  • Vendor claims 90%+ extraction accuracy
  • Claims broader PA lifecycle capabilities
  • 2–3 month implementation
Considerations
  • Payer integration approach unclear
  • File assembly / PDF manipulation untested
  • Trained-model paradigm (static between updates)
  • Pricing may not justify pilot at current volume
B

Salesforce Agentforce

Native CRM automation
Strengths
  • All data already in Salesforce
  • No new system to integrate
  • Familiar to Lumara IT team
  • Could handle routing/tasks natively
Considerations
  • Agentforce is very new and unproven
  • Salesforce historically oversells AI
  • Limited to Salesforce-internal operations
  • Won't handle Adobe, email, or file workflows
C

Custom AI-native build

Production application leveraging foundation models
Strengths
  • Built specifically for the PA workflow
  • Lumara owns the application and source code
  • PHI never leaves Lumara's cloud tenant
  • Workflow engine extensible to future processes
Considerations
  • Requires workflow mapping upfront (the PDD)
  • Extraction accuracy validated against real documents
  • Internal capability or retainer to maintain
  • Build cycle vs. drop-in vendor product
Our recommendation: pressure-test the vendor, demo Agentforce, scope a custom build — in parallel. Let evidence pick the winner.
The platform question

Build vs. Buy. In an AI-native world.

Both vendors and custom builds can use foundation models. The difference is who controls the workflow, the roadmap, and the data perimeter.

Build (AI-native)

Your application. Your roadmap. Your IP. Your data perimeter. Foundation models are an interchangeable substrate underneath.

As the substrate improves — and it improves rapidly — your application gets smarter without a vendor upgrade cycle.

Buy

Vendor's product. Vendor's roadmap. Vendor's pricing model. Vendor's data perimeter. The vendor decides when AI capability improves.

Whether they use a trained model or a foundation model internally is a tactical detail. The strategic limit is the same: you're buying their product, not building yours.

The Lumara onramp

A workflow engine you extend. Not a vendor roadmap you lobby for.

If the custom build approach succeeds for PA intake, Lumara has a workflow engine that extends to anywhere a human is touching Salesforce. Appeal packet assembly. Clinical doc review. Field force enablement. Account & territory planning. That's a fundamentally different ROI calculus than buying a point solution.

Beyond efficiency

Intake is Phase 1. Here's where the real leverage is.

The phased roadmap below shows how PA intake unlocks a broader automation capability across the PA lifecycle.

Phase 2

Authorization & appeal support

Auto-populate insurance fields from structured intake data. Build appeal packet assembly tools. Create payer-specific playbook templates from historical cases.

Phase 3

Appeals process orchestration

40+ step appeals processes that span multiple systems with multi-week follow-up cadences. Automate cadence management, portal monitoring, and document coordination across payer portals and external counsel.

Phase 4

Denial intelligence

Build a structured denial pattern database tracking payer-specific denial reasons, successful appeal approaches, and outcomes. Transform reactive appeals into strategic ones.

Strategic

Commercial payer expansion

As Lumara expands commercial coverage and patient volume grows, payer-specific complexity multiplies. The automation strategy must scale for this.

Suggested next steps

A two-week window to commit.

Each path needs evidence. Each evaluation has a defined owner and a clear endpoint.

1

Continue domain-vendor evaluation

Pressure-test scope. Can they handle file assembly, naming conventions, PDF manipulation? Or just intake/extraction/Salesforce filing? Clarify pricing and implementation timeline.

Genflare + Lumara
2

Cloud infrastructure & IT discussion

Working session with Lumara IT to align on cloud deployment, tenant configuration, HIPAA BAA requirements, and technical readiness for a pilot environment.

Genflare + Lumara IT
3

Intake workflow pilot design

Map the specific intake workflows in enough detail to design a pilot for one or both solution paths. Define success metrics. Outputs a Program Design Document with user stories and acceptance criteria.

Genflare
4

Salesforce Agentforce assessment

Quick investigation into Agentforce capabilities for PA workflow automation. Separate the hype from reality. Determine if it's a viable third path or a complementary tool.

Genflare
5

Review the full analysis

Detailed workflow analysis available for the team to review, comment, and flag anything we missed or got wrong. Becomes the living reference for the project.

Lumara PA team