Scaling Care Navigation: AI Workflow Automation & Knowledge Management
Well · Product Management Intern · Sep 2025 – Present · Chapel Hill, NC
The Challenge
As Well's digital health platform scaled, clinical and non-clinical support teams were hitting bottlenecks. Fragmented internal documentation made it hard for guides to quickly resolve member questions, and there was no standardized way to expose the platform's capabilities to AI-powered tools. At the same time, members increasingly expected to interact with their health data through AI assistants — and we had no infrastructure for that.
What I Did
AI workflow architecture
I authored two product requirements documents: one for an internal Model Context Protocol (MCP) server — the infrastructure that would let AI agents safely interact with the platform — and one for an AI-powered help center that would surface contextual answers to members in real time. Both involved mapping complex conversational workflows: when a member asks about benefits vs. when they need clinical triage, what gets automated vs. what requires a human handoff. I defined strict AI-to-human escalation boundaries to protect patient safety, and coordinated across engineering, clinical, and QA teams. The core design insight was 'build once, deploy everywhere' — one MCP server powering multiple AI surfaces rather than building custom integrations for each.
Knowledge management & cost deflection
I evaluated knowledge management platforms — conducting a competitive analysis of Salesforce Knowledge, Zendesk Guide, and GitBook against our core requirements: time-to-market, AI-powered search, actionable analytics, and brand fit. I built a financial model using real support volume data and demonstrated that even a conservative ticket deflection rate would yield significant ROI, freeing the team to focus on complex, high-touch member issues.
Agile execution
I wrote and maintained a backlog of user stories spanning both product workstreams, and partnered daily with engineering to keep cross-functional teams aligned — translating clinical needs into technical specs and keeping everyone building toward the same goal.
Portfolio Artifacts
MCP Server PRD
RedactedTechnical product requirements with AI-human handoff logic
AI Help Center PRD
RedactedMember-facing AI assistant product spec
Platform Evaluation
Competitive analysis with cost-benefit framework
Financial Impact Model
Support volume analysis and ROI projection
Impact
What I Learned
"The best product work I did here wasn't the PRD — it was learning to translate between clinical language and engineering language until everyone was building the same thing."