How and when Rezi escalates to you
Rezi hands off to a human for emergencies, unknown questions, and unhappy guests. Here’s how to control it.
The point of Rezi is to handle the routine so you only get pulled in when it matters. Escalation is how it knows when to tap you on the shoulder.
It helps to think of escalation as a feature, not a failure. An AI that never escalated would either be guessing at answers it doesn’t have or stonewalling guests who genuinely need a person, both worse than a clean handoff. The design goal is precision: every emergency and every stumped conversation reaches you fast, and nothing else does.
What triggers an escalation
- Emergency keywords: water leak, no heat, smoke, gas, lockout, and anything else you define
- Unknown questions: if the answer isn’t in the knowledge base, Rezi escalates instead of guessing
- Frustration: repeated complaints or angry messages get routed to a human
- Explicit requests: when a guest asks to “speak to a person”
The four triggers behave differently on purpose. Emergencies escalate instantly and unconditionally, no business-hours rule overrides them. Unknown questions escalate after Rezi tells the guest it’s checking with the manager, so the guest is never left hanging. Frustration detection watches the trajectory of a conversation, repeated complaints, all-caps, sharp language, and pulls a human in before things boil over. And an explicit request for a person is always honored immediately; the AI never argues a guest into staying with it.
How you get notified
Escalations reach you through the channels you configure in notifications, typically a text and/or email to the on-call contact. The full conversation is attached so you have context immediately.
What the guest experiences during a handoff matters just as much. Rezi tells them a human has been notified and sets an honest expectation, it doesn’t pretend you’ll reply in seconds at 3am. When you do step into the conversation, you reply from the dashboard (or by responding to the alert) and the guest sees the answer in the same thread, no new number, no “please call this other line.” Rezi steps aside while you’re engaged and resumes when you’re done.
Point escalations at a monitored number
Set your escalation contact to a phone you actually watch, especially after hours. For emergencies, configure a dedicated emergency flow.
Tuning sensitivity
You can adjust sensitivity and add custom keywords in Settings. Test your emergency keywords as part of the go-live checklist.
Expect to tune in both directions during the first weeks. Too many escalations usually means knowledge gaps, each “unknown question” handoff names a fact to add, see improving answer accuracy. Escalations that feel like false alarms usually mean an over-broad keyword; “leak” alone will catch “my faucet drips a little,” so prefer phrases like “water leak” and let the AI’s own urgency detection cover the rest. The reducing escalations playbook walks the whole loop.
Can different issues escalate to different people?
What does the guest see while waiting for me?
What if I miss an escalation?
Can I turn escalations off entirely?
Was this article helpful?