Today
Care-navigation transcripts (TalkDesk today, Zendesk after cutover) contain PHI. The AI Governance Board blocks AI from seeing them, so QA, summarization, and member-facing AI live downstream of a manual review wall. Manuel's QA project is the canonical blocked case.
How the workflow helps
Tray with Merlin Guardian masks PHI before transcripts reach the model, exposes only de-identified text to AI tools, and re-applies the original context on return — bringing care-navigation transcripts inside the Governance Board's PHI rules and unblocking QA, summarization, and member-facing AI downstream.
Value inputs to capture
- How many care-navigation transcripts arrive per month?
- What's the current manual QA cost per transcript?
- Which downstream AI projects are blocked on this gate today?
Add volume, time saved, and loaded cost to create a draft estimate. Until then, no impact number is shown.