Overview

CO

Proposed 90-Day Plan for the Chief AI Officer Role

Prepared by Joe Scanlin

North Star

Make RealMDai the most trustworthy way to turn patient context into a clinically usable plan.

Great AI

Structured reasoning + evidence retrieval + reliable outputs.

Great Workflow

Fast, safe review and signoff for MDs.

Great Auditability

Traceability of why recommended, what changed, what prescribed.

90-Day Plan Timeline

30

Days 0-30: Safety & Quality Foundation

Map clinical pipeline, define risk taxonomy, build evaluation harness, ship MD workflow v1, establish audit trail.

60

Days 31-60: Sharpen & Speed Up

Upgrade to structured clinical plans, cut MD review time by 50%, close learning loop, start integrations.

90

Days 61-90: Scale Responsibly

Scale doctor-verified workflow, build operational reliability, own 2-3 clinical verticals, establish team and cadence.

Days 0-30

Safety + Quality Foundation

Strategic Goals

  • Map the clinical pipeline. Document the complete flow from patient input to final MD decision. Identify every failure mode.
  • Define safety boundaries. Establish clinical risk taxonomy, hard stops, and escalation rules. Know what we will never do.
  • Build evaluation harness. Create automated testing for guideline adherence, contraindications, red flags, hallucinations, and clarity.
  • Ship MD workflow v1. Efficient review UI that captures structured reasons for every edit.
  • Establish audit trail. Immutable log for every encounter: inputs, outputs, checks, edits, final plan.

Key Deliverables

InitiativeOwnerDueStatus
Clinical Pipeline MapCAIO + Clinical LeadDay 7pending
Clinical Risk TaxonomyCAIO + Clinical AdvisorDay 14pending
Hard Stops and Escalation RulesCAIO + Legal/ComplianceDay 14pending
Evaluation Harness in CIML EngineersDay 21pending
MD Workflow v1Product + EngineeringDay 28pending
Audit Trail per EncounterEngineering + ComplianceDay 30pending

Risks & Mitigations

Model hallucinating dosage unitshigh
Mitigation: Hardcoded regex guardrails on all numerical dosage outputs.
PHI leakage in logshigh
Mitigation: PII scrubbing middleware implemented before logging layer.
Latency creating MD frictionmedium
Mitigation: Optimistic UI updates + streaming response integration.

Days 31-60

Sharpen & Speed Up

Strategic Goals

  • Upgrade to structured clinical plans. Move from chat responses to structured outputs: problem list, differential, tests, meds with dosing and contraindications.
  • Cut MD review time by 50%. Better UX: summaries, highlights, templated edits, keyboard shortcuts.
  • Close the learning loop. Turn MD edits into policy, prompt, and model improvements systematically.
  • Start integrations. Staged approach to EHR, labs, and eRx. Start minimal, prove value, then scale.

Key Deliverables

InitiativeOwnerDueStatus
Structured Plan OutputML Engineers + ClinicalDay 45pending
Retrieval + Citations LoopML EngineersDay 50pending
50% Review Time ReductionProduct + EngineeringDay 55pending
Learning Loop PipelineML Engineers + ClinicalDay 55pending
Near-Miss Tracking SystemEngineering + ClinicalDay 50pending
Integration MilestonesBackend EngineeringDay 60pending

Risks & Mitigations

Retrieval quality issueshigh
Mitigation: Curated, versioned guideline database. Human review of retrieval results. Confidence thresholds.
Learning loop introducing biasmedium
Mitigation: Aggregate across multiple MDs before making changes. A/B test prompt and model updates.
Alert fatigue from too many flagsmedium
Mitigation: High precision thresholding. Only surface actionable items.

System Architecture: The Learning Loop

Clinical Frontline

Structured Edits241

MD modifications captured as structured diffs (e.g., "Drug A -> Drug B").

Vector Store

Clusters Found3

Edits automatically clustered by semantic similarity (e.g., "Antibiotic Resistance").

Policy Council

Prompt Updates

Clinical lead reviews cluster. Updates system prompt: "Prioritize local resistance patterns."

Model Registry

v1.2-canary

New version deployed to 5% traffic. Evaluation harness confirms +4% accuracy.

Days 61-90

Scale Responsibly

Strategic Goals

  • Scale doctor-verified workflow. Production-ready verification: AI chat to structured encounter to MD sign-off with audit packet.
  • Build operational reliability. Monitoring, alerting, canary releases, automatic rollback triggers.
  • Own 2-3 clinical verticals. Deep expertise in high-volume, guideline-rich areas. Measurable advantage over alternatives.
  • Establish team and cadence. AI org plan, hiring, and operating rhythm for sustained clinical AI excellence.

Key Deliverables

InitiativeOwnerDueStatus
Scalable Verification WorkflowProduct + EngineeringDay 75pending
Production Monitoring DashboardEngineering + ML OpsDay 70pending
Canary Release SystemML Ops + EngineeringDay 75pending
2-3 Focused VerticalsCAIO + Clinical + ProductDay 85pending
AI Org PlanCAIO + CEODay 80pending
Operating Rhythm EstablishedCAIODay 90pending

Risks & Mitigations

Scaling too fasthigh
Mitigation: Explicit scaling gates tied to safety metrics. No volume increase until thresholds met.
Vertical focus wrongmedium
Mitigation: Data-driven selection based on current case mix. Willingness to pivot based on early results.
Key person dependenciesmedium
Mitigation: Document everything. Cross-train team members. Build redundancy into operating rhythm.

Metrics & Safety Dashboard

Live Preview
Filters:

Safety Event Rate

0.04%-12%

MD Overturn Rate

8.2%-2.1%

Avg Time to Review

1m 42s-15s

Escalation Accuracy

94.5%+1.2%

Volume vs. MD Intervention Rate

Review Time Distribution

Workflow Funnel (Daily)

Model Release History

v2.1.0
2026-01-15
full
v2.2.0
2026-02-01
full
v2.3.0
2026-02-15
full
v2.4.0
2026-03-01
full
v2.5.0
2026-03-10
staged
50%
v2.6.0
2026-03-20
canary
5%

Workflow Visualization

Interactive model showing data flow and verification steps.

Click each step to explore the workflow

Patient Input
AI Synthesis
MD Review
Final Plan

System Logic: Step 1

Ingest unstructured audio/text. Vectorize historical context. Retrieve relevant guidelines via RAG.

Data PreviewRead Only
45yo F, 3d cough, fever 101.2F, fatigue.

Recent "Near Miss" Events

Medication2026-02-14

Model suggested Amoxicillin for patient with Penicillin allergy listed in structured EHR field but not note.

Resolution: Context window expanded to include structured allergy list.

Guideline2026-02-20

Screening mammogram recommended at age 35 without family history.

Resolution: RAG pipeline updated with USPSTF 2026 guidelines.

Documentation2026-03-01

Hallucinated "non-smoker" status when field was empty.

Resolution: Added "unknown" token for missing fields.

Workflow2026-03-05

Plan generated before lab results returned.

Resolution: Added state check for pending results.

Dosing2026-03-08

Metformin 1000mg BID recommended without noting patient eGFR of 35.

Resolution: Mandatory eGFR check for metformin added.

Citation2026-03-12

Model cited non-existent guideline "AHA 2025 Headache Protocol".

Resolution: Citation verification layer added before output.

Team & Cadence

AI Team Hiring Plan

RoleCountTimingPriority
Chief AI Officer1Day 1Critical
ML Engineer1Days 1-30Critical
Software Engineer1Days 1-30Critical
Clinical Advisor (Part-time MD)1Days 1-30Critical
Total Headcount (90 days)4 roles

Operating Rhythm

Daily Clinical Review

Review flagged encounters, near-misses from past 24h

Daily, 15 min
Participants: CAIO, Clinical Lead, On-call MD

Weekly Safety Standup

Review safety metrics, near-miss trends, model performance

Weekly, 30 min
Participants: Full AI team, Clinical advisors

Bi-weekly Model Review

Evaluate model updates, plan releases, review learning loop

Every 2 weeks, 1 hour
Participants: ML Engineers, Clinical team, Product

Monthly Trust Report

Comprehensive safety/quality report for leadership and board

Monthly, prepared async
Participants: CAIO authors, Exec team reviews

Quarterly Clinical Board

Strategic review of clinical AI direction, major decisions

Quarterly, 2 hours
Participants: CAIO, CEO, CMO, External advisors

Reporting Structure

CEO
Chief AI Officer
ML Engineers
Clinical Lead
Product
Dotted line:Clinical Advisor (MD) reports to CMO with strong collaboration with CAIO

What Could Go Wrong & Contingencies

ScenarioContingency Plan
MD adoption is slower than expectedPivot to async review model. Reduce friction with mobile-first interface. Offer incentives for early adopters.
Model accuracy plateaus below targetNarrow scope to highest-confidence conditions. Invest in domain-specific fine-tuning. Partner with academic medical center for data.
Regulatory pushback on AI-assisted carePosition as "clinical decision support" not "diagnosis." Proactive engagement with state medical boards. Robust audit trail as compliance asset.
Key hire falls throughMaintain relationships with 2-3 backup candidates. Leverage fractional/consulting talent for interim coverage. Adjust timeline if needed.

Competitive Context

The clinical AI space is crowded but undifferentiated. Most competitors focus on either pure AI (no human oversight) or pure telehealth (no AI leverage). RealMDai's "doctor-verified AI" positioning is a genuine wedge.

Pure AI Players

Fast but risky. Regulatory exposure. Trust deficit with patients and providers.

Traditional Telehealth

Slow and expensive. MD bottleneck limits scale. No AI leverage on efficiency.

RealMDai (Our Position)

AI speed + MD trust + audit trail. Scalable and defensible. Regulatory-ready.