AI & knowledge base

What the AI can and can’t do

An honest look at Rezi’s strengths and its limits, so you know exactly what to expect.

Updated June 9, 2026 3 min read

Rezi is excellent at the high-volume, repetitive communication that eats your day. It’s deliberately conservative about anything that needs your judgment.

We’d rather you know the boundaries on day one than discover them mid-crisis. The short version: Rezi is a phenomenal first responder and a deliberately limited decision-maker. Both halves are by design, the same conservatism that stops it from inventing an answer also stops it from making calls that are rightfully yours.

What it does well

  • Answers factual questions from your knowledge base
  • Recognizes emergencies and alerts you right away with the details
  • Quotes availability and nightly rates from your synced calendar
  • Handles many conversations at once, instantly, in any language

The pattern behind that list: Rezi excels wherever the right response is determined by facts you’ve provided plus rules you’ve set. “What’s the WiFi?” has one right answer. “A guest reports no heat in January” has a defined playbook, reassure the guest, classify as urgent, and alert you right away so you can act. Within that territory it’s faster, more patient, and more consistent than any human could be at 3am across forty simultaneous conversations.

What it intentionally won’t do

  • Make up facts it doesn’t have, it escalates instead
  • Make legal or major financial decisions on its own
  • Override an emergency, those always reach a human
  • Change your pricing or issue refunds without your setup and approval

Each refusal protects you. Invented facts erode the trust the whole system depends on. Legal matters carry liability that belongs with a human who knows the situation. Emergencies need an accountable person, full stop. And money, pricing changes, refunds, fee waivers, moves only along rails you’ve explicitly laid: Rezi can quote the rates and fees you set, but it never negotiates or improvises financial terms.

The gray areas

Some situations sit between “clearly automatable” and “clearly human.” A guest disputing a late fee gets the policy from your knowledge base, but if they push back, that’s a judgment call and Rezi hands it to you. A guest asking to bend a house rule (“can we check out two hours late just this once?”) gets the rule as written plus an escalation if they press, unless you’ve put the flexibility into the knowledge base itself (“late checkout until 1pm is available for $40 when the calendar allows”). The more of your judgment you encode as rules, the more of the gray area Rezi can cover, but the default in ambiguity is always to involve you.

You’re always in control

You can jump into any conversation, and you decide which actions Rezi takes automatically versus which need your approval.

Setting your own boundaries

Several limits are dials, not walls. Escalation sensitivity and emergency keywords are yours to tune, as is how much of your judgment you encode as rules in the knowledge base. A sensible arc: start conservative, watch the Interactions view until the handling earns your trust, then loosen the dials where you’re comfortable. Most hosts end up far more automated than they expected, on their own schedule.

Will the AI ever guess when it’s not sure?
No. Uncertainty triggers either a clarifying question to the guest or an escalation to you. A confident wrong answer is the one failure mode the system is built to avoid.
Can Rezi handle a hostile or abusive guest?
It stays calm and professional, doesn’t take bait, and routes genuinely hostile conversations to you via the frustration trigger. It won’t conduct disputes on your behalf.
Does it give legal advice to guests?
No. Questions about legal rights or disputes are escalated. It can state your written policies, since those are your facts, but it won’t interpret law.
What if a guest tries to trick the AI?
The grounding design is the defense: Rezi can only assert what’s in your knowledge base, so a guest can’t talk it into a discount or a fake policy. Manipulative pushes tend to end in an escalation, not a concession.

Was this article helpful?

Related articles