How Suzy identifies herself
The first message a guest gets identifies the property and, depending on your settings, introduces Suzy by name as the property's AI assistant. This isn't optional dressing you can turn off to make the interaction feel more human than it is: US carrier rules for business text messaging (the same A2P 10DLC framework that governs whether your texts get delivered at all) and general TCPA guidance both push toward clear identification, and pretending to be a person when you're not is the kind of thing that erodes trust the moment a guest figures it out anyway.
The honest framing has turned out to work better than the alternative in practice: guests aren't offended that a fast, accurate answer came from software. What frustrates guests is a slow answer, a wrong answer, or an answer that ignores what they actually asked. Identification solves a legal requirement and, as a side effect, sets an accurate expectation the interaction then meets or beats.
What guests actually want at 11pm
Think about what an overnight guest question actually is: the WiFi password isn't working, what time is checkout, is there a vending machine, can I get a late checkout. None of these require empathy or a human touch to answer well. They require a fast, correct answer. A tired traveler at 11pm generally isn't looking for a conversation, they're looking for the answer so they can put the phone down and go to sleep.
For the smaller number of situations that do need a human touch, a genuine complaint, something confusing about the property, an emotional situation, Suzy is built to recognize that it's out of its depth and hand off, rather than trying to talk a frustrated guest through it with scripted empathy. That handoff is the actual safeguard for guest experience, not hiding the fact that an AI is involved.
Why disclosure isn't a downside
A guest who knows they're texting an assistant and gets a correct, instant answer generally rates that interaction well. A guest who suspects they were talking to a person, gets a wrong answer, and later learns it was automated feels misled twice. Clear identification up front avoids that second, worse outcome.
When it stops being the AI
The handoff isn't a vague promise, it's a specific mechanical rule. If Suzy's confidence in an answer falls below a set threshold, she tells the guest directly that she'll check with the front desk, and the question routes to the property manager's phone. The guest isn't left in a loop with a chatbot that keeps repeating itself; they get a clear, honest "let me get you a real answer" and a human follows up.
Real emergencies never touch this judgment call at all. A fixed set of crisis keywords, fire, a medical emergency, a break-in, and similar, triggers an immediate alert to the property manager through a hardcoded rule that runs before the AI even processes the message. This is deliberately not left to the AI's discretion in the moment.