AI vs Human Support: When to Escalate

The first question every AI-curious support team asks is the wrong one: "Can AI handle this?" The right question is: "When should AI stop handling this?"

The 70/30 baseline

In our data across hundreds of workspaces using Connect, the steady-state mix is roughly 70% of conversations fully resolved by AI and 30% routed to a human at some point. Teams that try to push the ratio higher with weaker escalation rules see CSAT crater within a week.

Five signals that should always trigger a human

1. Customer explicitly asks for a human.
2. AI confidence drops below your threshold (we recommend 0.6 as a floor).
3. The topic touches money, accounts, legal, or safety.
4. The customer has messaged 3+ times without resolution.
5. Sentiment turns negative — frustration is a sign you're losing them.

Why "AI assist" beats "AI replace"

The teams who win don't replace agents with AI. They give every agent an AI co-pilot. Drafting replies, summarizing context, suggesting articles — agent throughput doubles without losing the human voice.

Operationalize it

Configure escalation rules in your help-desk software (Connect ships with these out of the box), train your team on when to override the AI, and review a sample of AI-only conversations weekly. The goal isn't to eliminate humans. It's to make humans available for the conversations that need them.