Automate Concierge Requests with a Custom AI System

Most resorts rely on a combination of a Property Management System (PMS) and a guest messaging platform like Kipsu or Whistle. These messaging tools are excellent for human-to-human communication, but their automation is typically limited to simple keyword triggers. They can route a message containing "spa" to the spa desk, but they cannot understand or act on a complex, multi-part request.

Consider a guest texting: "Hi, can you book us a cabana for tomorrow at 2 PM and also arrange a late checkout for our room, 405?" The keyword-based bot fails immediately. It might send a link to book a cabana but will completely ignore the late checkout request. A human concierge must intervene, open the amenity booking software to check cabana availability, then open the PMS (often a legacy system like Oracle OPERA) to check room 405's status and availability for late checkout. This context switching between two disconnected systems takes 5-10 minutes and is prone to error.

The structural problem is that these tools are data silos. The messaging platform has no real-time, stateful awareness of PMS data like room inventory or guest folio details. The PMS was built as a system of record, not a real-time conversational engine. Attempts to connect them with generic tools often fail due to the restrictive APIs of older PMS platforms or the lack of sophisticated logic needed to orchestrate a multi-step guest request.

The engagement would begin with an audit of your current PMS and guest communication channels. Syntora would analyze your PMS's API documentation to map out how to access reservation data, guest profiles, and amenity inventory. We'd also review a log of the 100 most frequent guest requests to define the initial scope for the AI agent.

The system's core would be a Python FastAPI service deployed on AWS Lambda for cost-effective, serverless scaling. This service uses the Claude API to parse the natural language of guest requests, identifying intents (e.g., 'book_amenity', 'request_late_checkout') and entities ('cabana', 'tomorrow 2 PM'). Pydantic models validate this extracted data before any action is taken. The service then communicates directly with your PMS API to check availability and with a Supabase Postgres database to log every interaction for auditing. A typical request would be fully processed in under 800 milliseconds.

The delivered system integrates into your existing guest messaging platform via webhook or can be attached to a dedicated phone number. Your concierge team gets a simple dashboard to monitor AI conversations and can take over any chat with a single click. You receive the full source code in your GitHub repository, complete with a runbook detailing how to update the AI's knowledge or add new request types.