shortrental.ai

AI modes: Shadow, Live, Escalation

When to use each AI mode, and how to level up the AI's trust over time.

Updated: Apr 8, 2026

Shortrental's AI operates in one of three modes per workspace. Here's when to use each.

Shadow Mode

What it does: The AI reads every incoming message and drafts a reply. Nothing is sent until you approve it.

Use it when:

  • You just signed up
  • You added a new language or market
  • You changed your brand tone
  • You're onboarding a new type of property

Move on when: You've approved ~50 consecutive drafts with minor or no edits.

Live Mode

What it does: The AI sends replies directly, 24/7, in under 30 seconds. Every action is logged and one-click reversible for 24 hours.

Use it when:

  • You trust the AI's tone and accuracy on 90%+ of common queries
  • You want to scale without hiring

What's still escalated to you:

  • Refund requests
  • Complaints
  • Damage / dispute handling
  • Complex legal or insurance questions
  • Any message matching your custom escalation keywords

Escalation Mode (automatic within Live)

The AI continuously classifies every message. When confidence drops below a threshold (default 80%) or the message matches an escalation keyword, the AI:

  1. Sends a holding reply: "Thanks for reaching out — I'm flagging this with the team and we'll be back to you shortly."
  2. Notifies the right team member via Slack, email or SMS
  3. Pre-fills the context: previous messages, booking details, suggested actions
  4. Logs the escalation for review

Configuring escalation

AI Assistant → Escalation rules

  • Keyword triggers (add your own, e.g. "legal", "fraud", "police")
  • Confidence threshold (default 0.8)
  • Recipient routing (who gets which type)
  • Escalation SLA (auto-bump after N minutes if unresponded)

Levelling up the AI

The AI improves continuously based on your edits. In AI Assistant → Knowledge, you can:

  • Add rules (e.g., "Always recommend Parking P2 for this building")
  • Add FAQ pairs (question → preferred answer)
  • Mark conversations as canonical training examples
  • Bulk-rate past replies

Changes take effect immediately for new messages.