AI human handoff: escalation rules for CSAT

Nothing ruins customer satisfaction faster than an AI bot that refuses to get out of the way — except an AI bot that hands you over to a human who immediately asks, "How can I help you today?" as if the previous ten messages never happened.

Good AI customer support is measured not only by the 60–80% of repetitive questions the agent resolves on its own, but by how cleanly it hands over the remaining 20–40%. Here is how to configure escalation rules in Threads so CSAT goes up instead of down.

The five escalation triggers every agent should have

In Threadbox, escalation rules run after every turn and evaluate five signals:

1. Explicit human request (tool call): Whenever a visitor asks to speak to a person, the model calls the built-in escalate_to_human tool immediately. Never force a customer to ask twice.
2. High-risk keywords: Phrases like chargeback, solicitor, damaged, ombudsman, leak or cancel account should bypass further AI troubleshooting and route straight to the right team.
3. Negative sentiment threshold: If the sentiment score drops below your threshold, pause the AI before frustration turns into a public review.
4. Repeated fallback count: If the AI uses its fallback answer twice in a row, it does not have the knowledge to help — escalate automatically.
5. Message-count ceiling: If a conversation exceeds N visitor turns without resolution, bring a person in.

What must happen at the moment of handoff

• Pause the AI immediately: Once escalated, the AI must stop replying so it never talks over your human agent.
• Attach a Sidekick summary and CRM context: In Threads, Sidekick summarises what the customer wants and what the AI already checked, alongside live order or account data from Shopify, Stripe, HubSpot or Salesforce.
• Start a business-hours response target: Assign a priority and start an SLA timer that respects your team's UK working hours and bank holidays.
• Reply on the customer's original channel: Whether the conversation started on the website widget, email or WhatsApp, your agent replies in Threads and the customer receives it on the same channel.

Closing the loop: turn escalations into new knowledge

Every escalation is a signal that either an action was missing or a help guide was unclear. When your agent resolves the thread in Threads, one click in Sidekick drafts a new Help Hub guide from the resolution so the AI agent can answer it next time. Explore Threads or test escalation flows in our live interactive demo.