Overview

This lesson is for admins and operations leads. It pulls together everything earlier in the curriculum into a single setup checklist, then walks through the tuning patterns that matter when you're running Conversations across a large operation.

If you only manage a handful of properties and a single agent, you can skim. If you're rolling Sparrow Intel out to a team handling thousands of conversations a week, read carefully.

Part 1: Setup checklist

Connect your channels

• PMS connected
• Direct Airbnb integration connected (recommended in addition to PMS-pass-through)
• Email integration set up (forwarding address or custom domain)
• Verified test conversations are arriving on each channel

Setup guides for each PMS live in the Getting Started collection. To send and receive from your own domain, see Configuring Custom Domain Email.

Add your team

• Each team member invited as a user
• Roles assigned appropriately
• Team members have logged in and configured their notification preferences

See Add an Additional User for the invite flow.

Configure conversation categories

• Reviewed the default 8 categories (Routine Inquiry, Check-in Logistics, Issue Report, Complaint, Booking Modification, Emergency, Billing, Compliment)
• Added any custom categories specific to your operation
• Removed any default categories that don't apply

Manage these in Conversation Settings (Conversation Categories).

Build your knowledge snippet library

• At least one snippet per high-frequency question (parking, wifi, check-in, late check-out, pet policy)
• Property-specific snippets for any property with non-standard amenities or rules
• Owners of snippet maintenance identified

Build and maintain these in the Knowledge Hub (guidebooks and knowledge snippets).

This is the single highest-leverage thing you can do for AI quality. A team that invests three days in writing 50 good snippets has dramatically better AI suggestions than a team with a dozen hastily-written ones.

Build your message templates

• Templates for the 5–10 highest-frequency outbound messages
• Templates use variables ({{GuestFirstName}}, {{PropertyName}}, etc.) so they don't read robotically
• Templates audited for stale information (phone numbers, addresses, policies)

Manage these on the Message Templates page.

Set up Rules

• Set up rules for your highest-value automation patterns — see Rules for Conversations for a starting example
• Test each rule with a label-only action before enabling real-world side effects

Configure Autopilot — carefully

• Pick a small, well-understood property group to start with — not your entire portfolio
• Set conservative thresholds (confidence 90, max risk Low, all quality signals required)
• Limit to safe categories (Routine Inquiry, Compliment)
• Designate a daily/weekly reviewer for the first month

Re-read Autopilot: Auto-Reply Safely at Scale in full before turning anything on.

Connect your task system

• Operations integration connected (Breezeway, PMS-native tasks, Asana, or webhook)
• Test task created from a real conversation, end-to-end
• Department mapping verified (cleaning issues land with the cleaning team, etc.)

See the Task Creation Guide.

Part 2: Tuning for high-volume operations

The above gets you running. The below gets you running well at scale.

Triage view discipline

In a high-volume inbox, the default "everything Open" view is not workable. Build saved views your team uses by default:

• Unassigned + Open — the daily intake queue
• Mine + Open — your own work
• Negative sentiment + Open + last 24h — the escalations queue
• Property group X + Open — for region- or brand-specific teams

Train every agent to start their day in one of these — never in the unfiltered firehose.

Assignment patterns

Two patterns scale; pick the one that fits:

Sparrow Intel doesn't pick a pattern for you — it gives you the assignment, label, and rule primitives to implement either one.

Knowledge snippet hygiene

The snippet library is a living thing. The bigger your operation, the more it drifts. A monthly cadence works for most teams:

• Audit for stale snippets (anything older than 6 months without a review)
• Look for duplicates created by different team members
• Identify gaps from past Autopilot misses or Chirp suggestion edits

A team running Autopilot at scale should have a designated snippet owner. Without one, drift wins.

Autopilot tuning rhythm

Once Autopilot is running, set a recurring review:

• Weekly for the first month — sample 20 Autopilot-handled conversations; flag any you'd have written differently
• Monthly thereafter — broader sample, look for patterns
• Whenever a metric moves — increased correction rate, increased held-for-review rate, sentiment trend changes

Adjust thresholds, categories, and snippets based on what you find. Autopilot quality is downstream of snippet quality — most "Autopilot got it wrong" issues are actually "snippet was missing or wrong."

Metrics to watch

Built-in reporting gives you the basics. For deep analytics, the webhook action lets you stream conversation events to whatever BI tool you use.

Onboarding new agents

Once Conversations is dialed in, you'll keep adding agents. A standard onboarding:

1. Read the Flight School Conversations track end-to-end (this curriculum)
2. Shadow a working agent for a half-day
3. Start with read-only access to the inbox; observe how senior agents handle a range of conversations
4. Switch to "reply but no Autopilot privileges" for the first week
5. Full access from week 2

For the team handling 200,000 reservations a year, an agent's ramp from "first day" to "full productivity" should take days, not weeks. The curriculum and the AI assistance both shorten that timeline if you let them.

When to ask for help

Before reaching out to support, ask yourself: was knowledge missing? If Chirp AI said something wrong or incomplete, there's a good chance a knowledge snippet would fix it — and any team member can add one in under a minute. More snippets means better suggestions immediately and a higher Autopilot reply rate over time as the knowledge base grows.

The in-product chat icon (bottom-right of the portal) reaches our support team. Things that are worth pinging us about:

• Persistent misclassification (Chirp keeps reading X as Y)
• Channel sync issues that don't resolve in a day
• Autopilot sending things that shouldn't have been sent (we want to know — this is how we improve)
• Integration questions where the docs aren't clear
• Feature requests — we read every one

Where to next

• Browse the rest of the help center for deeper references on individual features
• Watch for additions to Flight School — the next track in development is Reputation Management (Reviews, Predictions, automated review responses)
• Use the in-product chat for anything specific to your account

Welcome to flying solo. ✈️