When does the AI escalate from monitoring?

Five conditions end monitoring and bring in your team. The AI escalates when the customer expresses frustration, when the issue requires actions beyond the available tools, when the same problem persists after three or more follow-ups with no progress, when the customer requests a human, or when it's unsure how to proceed. Until one of those is true, automated check-ins continue.

The customer-driven ones

Frustration and 'can I talk to a person' are hard stops. The point of monitoring is patience, not stonewalling, so a frustrated shopper or an explicit request for a human means staff own the thread promptly, not after another silent check-in.

The capability-driven ones

Some cases need refunds, exceptions, or workflows your configuration doesn't expose to the AI. Those escalate rather than stall. Uncertainty works the same way: when the AI is unsure how to proceed, it hands off instead of guessing or looping indefinitely. If you see frequent beyond-the-tools escalations, look at your tool configuration before blaming difficult customers. It's often a gap you can close.

The stuck-loop cap

Three or more follow-ups with no meaningful change triggers staff involvement on its own. A shipment that never updates shouldn't be monitored forever; the cap protects the customer from an endless loop.

After the handoff

Escalated conversations land in your normal workflow: the inbox, plus notifications such as Slack if configured. Staff reply from the dashboard, adjust the case, and can optionally return it to monitoring later if your process allows. Monitoring alone won't keep running past these guardrails once an escalation condition is met.