Current Workflow: Support agent manually categorizes emails, finds approved guidance, drafts replies, requests approval for sensitive cases, sends responses, and records outcomes. AI Opportunity: Classify emails and draft replies for human approval. Available Resources: Owner, one agent, 50 anonymized emails, spreadsheet, 14 days, limited review time, and manual fallback. Risk Boundaries: No confidential data in unapproved tools, no autonomous sending, no refund decisions, and human review of every reply. Success Measure: Reduce measured handling time by 25 percent, maintain quality of four out of five, record zero unsupported claims, and zero privacy incidents.