AI agents for customer support that don't make customers angry

How to deploy an AI support agent that actually helps — scoping, escalation design, tone, and the metrics that prove it's working.

The bar is higher than chat

A bad AI agent loses customers faster than no agent at all. The bar isn't "answers questions" — it's "leaves the customer feeling respected."

Scope before you build

What categories will it handle? (account, billing, how-to, technical)
What will it never touch? (refunds above $X, security issues, churn risk)
How does it escalate, and how fast?

Escalation is the product

Customers don't mind talking to a bot if escalation is one click, fast, and arrives with full context. Every conversation should have a visible "talk to a human" button from message one.

Tone matters

Short answers beat thorough ones for routine questions
No "I'm sorry to hear that" theatre — acknowledge briefly and move to the fix
Match your brand voice, not the default ChatGPT cadence

What to track

Containment rate, but only on resolved-positively conversations
Time to human handoff when escalated
CSAT delta between bot-handled and human-handled tickets
Top 10 questions the bot got wrong this week — review every Monday