Company Context
You are the PM for CarePath AI, a Series D healthcare SaaS company that sells an AI-assisted clinical workflow product to mid-to-large US health systems. CarePath AI integrates into Epic and Cerner and is used by 38 hospital systems, covering ~22,000 clinicians and ~9.5M patient visits/year. Revenue is subscription-based ($120K–$900K per hospital/year) with renewals heavily tied to clinician adoption and measurable quality outcomes.
CarePath AI’s flagship feature, Clinical Copilot, summarizes the chart and drafts parts of the note (HPI, assessment, plan) and suggests next steps (labs, imaging, referrals) based on the patient’s record. The company is preparing to expand Copilot into inpatient medicine (higher acuity, more complex teams) after early success in ambulatory settings.
User / Market Scenario
Primary personas
| Persona | Share of usage | Primary job-to-be-done | What “aligned with patient care” means to them |
|---|
| Hospitalist (inpatient) | 35% | Manage many patients safely under time pressure | Fewer missed diagnoses, fewer medication errors, smoother handoffs |
| Bedside RN | 25% | Execute orders, monitor changes, escalate quickly | Clear, actionable plans; fewer contradictory orders; less rework |
| Clinical Pharmacist | 10% | Prevent adverse drug events | Accurate med reconciliation, interaction checks, renal dosing |
| Quality & Safety Lead | 15% | Improve quality metrics and reduce harm | Lower readmissions, lower sepsis mortality, fewer safety events |
| Patient / Caregiver (indirect) | — | Understand plan and feel safe | Clear explanations, fewer delays, fewer avoidable complications |
Competitive landscape
- Nuance DAX / Microsoft: strong ambient documentation; less focused on care-gap closure.
- Abridge / Suki: clinician-facing note generation; variable integration depth.
- Epic native AI features: best workflow integration; hospitals prefer fewer vendors.
CarePath’s differentiation is “documentation + decision support” (not just scribing). That differentiation also increases safety and regulatory scrutiny.
Problem / Opportunity
In a 12-week pilot across 3 hospitals (inpatient medicine), Copilot increased note completion speed by 18% and improved clinician satisfaction (CSAT 4.2 → 4.5/5). However, the pilot surfaced alignment issues:
- Safety drift: In 2.1% of encounters, Copilot suggested a next step that conflicted with local guidelines (e.g., antibiotic choice not matching antibiogram, anticoag dosing without renal adjustment).
- Objective mismatch: Clinicians report Copilot sometimes optimizes for “complete documentation” rather than “best next action,” adding low-value tests. Ordering rate increased 6% with no clear outcome improvement.
- Workflow friction: Nurses report that AI-generated plans are occasionally ambiguous (“monitor closely”), leading to more pages/calls. RN-reported “clarification needed” flags rose from 7% → 11%.
- Trust and accountability: Risk management is concerned about who is responsible when AI suggestions contribute to harm. The legal team is worried about discoverability of AI drafts.
The CEO has set a 2-quarter goal: expand to 10 additional hospitals while maintaining a safety bar that Quality & Safety leaders will sign off on.
Your Task (what you will be evaluated on)
As the PM, outline how you would ensure CarePath AI remains aligned with patient care objectives as it scales.
- Define “alignment” for this product: whose objectives matter, how you translate them into product requirements, and how you avoid optimizing for the wrong proxy.
- Choose a target segment and use case for the next 6 months (e.g., sepsis early management, med reconciliation, discharge planning). Explain why.
- Propose an MVP and prioritization: what you would build/change in Copilot (product + workflow + model behavior) to improve alignment.
- Design the measurement system: metrics, guardrails, and how you would evaluate success without waiting years for long-term outcomes.
- Identify key risks and mitigations: clinical safety, bias/fairness, regulatory, and operational risks.
Constraints
- Timeline: MVP must ship in 10 weeks to support the next sales cycle.
- Team: 1 PM, 6 engineers, 2 applied ML engineers, 1 clinical informaticist (0.5 FTE), 1 designer.
- Data & integration: Limited real-time signals; most hospitals provide HL7/FHIR feeds with 15–30 min latency. Order entry is in the EHR; Copilot can only suggest, not place orders.
- Regulatory / compliance: Must meet HIPAA; hospitals require audit logs. Some features may be considered SaMD depending on claims.
- Safety requirement: Quality leaders demand <0.5% guideline-conflicting suggestions in the top 3 recommended actions before expanding beyond pilot sites.
