Probably AGI
Concept proposal prepared for UnitedHealth Group + OptumNot affiliated with or endorsed by UnitedHealth Group
The opportunity beyond code intelligence

Build the evidence fabric for a modern health system.

UnitedHealth Group can connect software, policy, AI models, operations, clinical research and human decisions without collapsing them into one ungoverned data lake—or one unchallengeable answer.

PROVENEVIDENCE
FABRIC
every claim keeps its lineage
CODE
POLICY
MODELS
OPERATIONS
RESEARCH
DECISIONS
01Separate authorities

Clinical, security, engineering and policy judgment remain distinct.

02Shared evidence

Each reviewer can inspect the same lineage and current horizon.

03Explicit uncertainty

Unknown, stale and unverified are product states—not footnotes.

01 / THE THESIS

The biggest opportunity is not another model.

It is a durable layer that tells every model, engineer, reviewer and executive what is known, why it is known, what changed and what would prove the decision wrong.
01CODE

services · contracts · dependencies

02POLICY

controls · standards · obligations

03MODELS

lineage · evaluation · uncertainty

04OPERATIONS

workflows · incidents · recovery

05RESEARCH

literature · cohorts · contradictions

06DECISIONS

human authority · receipts · outcomes

02 / OPPORTUNITY MAP

Start credible. Expand audaciously.

The labels matter. Some capabilities are ready for a bounded pilot; some require governed product work; the biggest ideas should earn their way into production.

01PILOT-READY

AI Review Board Flight Recorder

A review packet for every consequential model change: source evidence, policy controls, uncertainty, test results, human overrides and downstream impacts.

VALUEFaster, more reproducible Responsible AI review
02FOUNDATION EXISTS

Living Enterprise Evidence Atlas

Continuously map code, policy, models, documentation and operational boundaries—then show which claims are current and which have drifted.

VALUEOne inspectable truth surface without one giant source of authority
03GOVERNED BUILD

Prior Authorization Evidence Chain

Connect policy language, implementation paths, model assistance, reviewer decisions and change history while preserving qualified human authority.

VALUEAuditability without automating clinical denials
04PHASE 2

FoxScience Evidence Workbench

Synthesize literature, aggregate cohorts, mechanisms, contradictions and candidate interventions into explicitly review-required evidence maps.

VALUEResearch acceleration with provenance and human review
05MOONSHOT

Critical Workflow Recovery Twin

Continuously rehearse synthetic claims, pharmacy and payment journeys against dependency changes and recovery plans—without touching production records.

VALUERecovery evidence before the next disruption
06MOONSHOT

Health-System Decision Memory

Preserve why a decision was made, the evidence available at the time, what later changed and which outcome falsified the original assumption.

VALUEAn enterprise that learns instead of merely logging
03 / MEET THE ESTATE WHERE IT IS

Evidence across the systems UHG already operates.

Public evidence indicates SAP is part of UHG/Optum finance and procurement workflows—including SAP, Ariba and Fieldglass—but exact present scope should be confirmed in discovery. The product opportunity is broader than any one platform.

SAP / Ariba / Fieldglass

Publicly indicated — validate scope

Cloud + data platforms

Connect through read-only evidence adapters

Clinical + administrative systems

Map boundaries before data movement

United AI Studio

Complement governance; do not replace it

Policies + standards

Bind implementation to obligation

04 / THE GOVERNING BOUNDARY

Earn trust before expanding authority.

The first deployment is read-only, no-PHI and bounded to engineering and documentation evidence. Health intelligence follows only after privacy controls, append-only review receipts and qualified human authority survive hostile testing.

NOAutonomous clinical denials
NODiagnosis or treatment authority
NOPHI in the initial PX pilot
YESHuman-review receipts and explicit escalation
YESUHG-controlled deployment and auditability
THE CONVERSATION TO START

Three questions for the room.

  1. 01Which code, policy or model estate consumes the most time proving what is actually true?
  2. 02Where does evidence break between engineering, operations and Responsible AI review?
  3. 03What would a six-week pilot need to prove for UHG to expand it?
THE ORDER MATTERS

Start with evidence. Earn the right to expand.

Begin with the PX pilot →

PROBAGI / SOURCE PROOF

The page keeps its evidence attached.

packaged

Engineering evidence fabric

The software-evidence foundation is framed from existing PX capabilities; UHG integration is not yet implemented.

px-uhg-evidence-fabric-v1
source-proven

Enterprise platform discovery leads

Public materials indicate role-specific or historical SAP, Ariba, Fieldglass, PeopleSoft, Concur, and OnBase usage; present enterprise scope remains a discovery question.

uhg-enterprise-platform-public-evidence-2026-07-28
held

Health-intelligence extensions

These are governed pilot and phase-two opportunities, not deployed UHG capabilities or autonomous clinical authority.

uhg-health-intelligence-opportunity-map-v1