Overview
The AI Messaging Agent is an AI-powered feature that helps property managers automate guest communication within Breezeway Messaging. It comes in two modes Copilot and Autopilot. These operate on the same underlying intelligence but differ in how much of the work the AI tries to does on its own.
The Agent automatically analyzes inbound guest messages, looks up relevant information from your Property Profile, Blueprint, Guide, and Elements, and either drafts a reply for a user to review (Copilot) or sends one directly to the guest when confidence is high enough (Autopilot). It also detects issues in conversations and surfaces Smart Tasks at the right moment, and learns from how users respond when it can't answer on its own.
Copilot vs. Autopilot
Copilot Mode: AI analyzes every inbound guest message and suggests a reply, a Smart Task, or both. You can then decide what to send. AI will learn automatically from how users respond through edited suggested replies or manually sent replies, especially helpful when AI doesn’t have any information to confident in it’s response.
Autopilot Mode: AI analyzes every inbound message associated with a reservation and, when confidence meets the threshold the customer has set, automatically replies to the guest. When confidence is too low, it falls back to Copilot behavior (suggested reply and a call to action to add knowledge).
💡You may choose to start in Copilot to help train the AI on content, then switch to Autopilot once you're confident and happy with how Copilot is generating suggested responses.
Copilot Mode
Copilot is the assisted layer of the AI Messaging Agent. It analyzes every inbound guest message, decides whether it has enough information to answer, and presents a suggested reply, a Smart Task, or both directly inside the Messaging conversation.
How Copilot Works
Message Analysis: When a new inbound guest message comes in, Copilot automatically analyzes it to determine whether the message is a question, contains a reportable issue, or both.
Context Gathering: The system pulls relevant information from the customer's account:
Property Profile (including custom fields)
Blueprint
Guide
Confidence Scoring: The AI determines how confident it is that it has the right information to answer - Low, Medium, High or Very High.
Suggested Reply: A suggested reply is displayed prominently in the conversation with Edit and Send actions. Source indicators show exactly where the information came from.
Smart Task Suggestion: If Copilot detects an issue in the guest's message (e.g. broken appliance, access issue, cleanliness concern), it surfaces a Smart Task call to action right in the conversation so you can escalate in 1-2 clicks. As we observe the creation of smart tasks through Copilot, we’re able to learn about what types of issues are tasks are typically created for from guest conversations.
AI Learning Loop: When confidence is Low and a user sends a manual reply, Copilot analyzes that response to learn from it. If a learning is captured, it's logged in the AI Activity Log and the outbound message is tagged so the user knows Copilot learned from it.
Note: Learnings will be readily available for Administrators to view within the AI Activity Log. These learnings can be undone/deleted if you do not want AI taking it into account moving forward.
Optimizing Copilot
Copilot's quality of answers generated will depend on the depth of information in Blueprint and throughout Property Profiles, as well as observed learnings that the AI has picked up on from your Messaging usage.
Manually Respond or Edit a Suggested Replies: When you write or edit a reply to a guest, AI picks up on meaningful patterns and applies them automatically next time a similar situation comes up.
Thorough Sources: The more information you have documented in Blueprint and in property profile the higher Copilot's confidence will be being able to answer inbound questions and issues from guests. When Copilot can't answer, simply reply like you normally would and have AI learn from the conversation from the action.
Blueprint: General information, policies, procedures, rules that are applicable across all properties.
Examples - Refund policies, Linens / What is provided, Requests and Modification information, Heating / Cooling Issues, Broken Appliance, etc.
Property Profile: More property specific information, traits, features.
Examples - Pool / Hot Tub + Instructions, Specific amenities, parking information specific to a property, Laundry, Fireplace, Coffee Maker, Sauna, House Rules (specific to this property)
Fill out Suggested Actions in AI Activity Log: We’ll capture more learnings from your usage over time and continue to learn, but you’ll want to make sure to promote important learnings to Blueprint and Property Profile as you see fit.
Use Custom Property Profile fields: Input property-specific details (e.g., Pool, Hot Tub, Roof Deck) help Copilot answer property-specific questions accurately.
Autopilot Mode
Autopilot is the fully automated layer of the AI Messaging Agent. When enabled, Autopilot replies to guest messages on your behalf within the settings and guardrails you have configured. It only sends a reply when its confidence meets or exceeds the threshold set in settings based on information it finds, and it only acts on messages from guest or owners with reservations (depends on settings).
When Autopilot can't confidently respond, it falls back to Copilot behavior: it leaves a suggested reply for you to edit or dismiss. As noted above, we'll observe and analyze your reply to see if there is a notable learning so that we don't make that same mistake next time.
How Autopilot Works
Reservation Check: Before doing anything, Autopilot checks whether the guest's phone number is associated with an active current reservation. If there's no reservation, Autopilot does not act on the inbound message.
Inbound Analysis: Autopilot scans Blueprint, Property Profile, Guide, and Elements for relevant information to answer the inquiry, similar to Copilot.
Confidence Check: The AI determines its confidence rate. If it is at or above the your configured threshold, Autopilot will proceed to send the message.
Send & Action: Autopilot sends the reply and applies the configured action — No Action or Mark as closed.
Fallback to Copilot: If confidence is below the threshold, no reply is sent. Instead, Copilot behavior takes over: a suggested reply is shown and you can proceed to manually answer so our AI can automatically learn for next time this comes up.
Conversation Marked as “AI Needs Information”: When this occurs in autopilot mode, the conversation will receive a purple “AI needs information” treatment. This treatment will stay in place until the guest receives a reply back. If the notification for “AI Needs information” is turned on within Settings → Notifications, you will receive an in-app nudge within the desktop version of Breezeway.
Optimizing Autopilot
Set a confidence threshold that matches your risk tolerance: Options are Medium, High, and Very High
Note: Our default for this is High
Use the schedule to control when Autopilot is active: Configure specific days and times when Autopilot is allowed to respond. Outside those windows, it stays in Copilot behavior.
Configure notifications: Turn on email, SMS, or in-app desktop notifications for when AI needs information and can't respond or when a Smart Task is suggested, so your team knows when human attention is needed.
Notifications
With Messaging access, you will be able to opt-in to receiving notifications for a few scenarios:
AI Needs Information: When AI isn't able to respond to a guest automatically when you have Autopilot enabled
AI Suggests a Task: When AI detects a message requires immediate action, the AI automatically flags this to you and notifies you with an auto-suggested Smart task. You can then click from that suggestion and we’ll generate a smart task with all the issue and conversation context for you to create within 1-2 clicks.
Settings & Guardrails
You can configure the AI Agent under Settings --> Messaging --> AI Agent.
Mode - Choose between Copilot and Autopilot
Default Tone - Choose from Professional, Friendly, Concise, Casual or Empathetic. The default selection is Friendly.
Autopilot Specific Settings
Contact Types - The default selection is Guests Only
Guests Only - When this option is used AI autopilot will only automatically respond to guest type reservations.
Owners Only - When this option is used AI autopilot will only automatically respond to owner type reservations.
Guests and Owners - When this option is used AI autopilot will only automatically respond to both guest and owner type reservations.
Confidence Rate - Medium, High or Very High. The default selection is High.
Action After Reply - The default selection is No action.
No action - When this option is used, if the AI agent in autopilot mode is able to respond, it won't take any additional actions.
Mark as closed - When this option is used if the AI agent in autopilot mode is able to respond, and no suggested smart tasks were present then it will automatically close out the conversation. The conversation will be re-opened if the guest responds.
Schedule - You can use this to if you only want Autopilot to run on specific defined days/times.
On/Off Levels: AI Agent can be toggled at the global, or property level.
Property level can be found in a property's settings.
Up next- learn more about the AI Activity Log here.