Introducing AI Triage: Semantic Cache with Safe Handoff

Support teams face a impossible tradeoff: respond instantly with AI, or respond accurately with humans. Most chatbots pick speed—and erode customer trust when they guess wrong.

EdgeDesk AI Triage uses Cloudflare Vectorize for semantic caching. When an incoming question matches a knowledge-base article with cosine similarity ≥ 0.95, the AI responds instantly without an LLM call.

Below that threshold, the ticket routes immediately to a human agent. The AI never fabricates an answer. This is a deliberate product decision: silence is better than a wrong answer.

When an agent sends their first reply, the bot is permanently disabled for that ticket until manually re-enabled. This follows the free-live-chat pattern: humans take over, AI steps aside.

Escalation also triggers on negative sentiment, payment or legal keywords, and two consecutive follow-up questions. Every escalation reason is recorded in the ticket timeline for QA.

AI Triage is available on Growth and Enterprise plans. Starter includes ticket management without AI generation.