Build a Custom AI for Guest Check-In and Concierge

Most hospitality businesses rely on their Property Management System's (PMS) built-in messaging. Tools within platforms like Cloudbeds or Mews can send templated messages triggered by events like a new booking or a date-based reminder. These are one-way announcements. They cannot understand or respond to a guest's question like, "Can I check in early?" This forces the front desk team to manually monitor and reply, interrupting their focus on in-person guests.

To fill this gap, some properties add generic website chatbots like Tidio or Intercom. These tools can answer basic, pre-programmed FAQs from a decision tree. For example, a guest asks about check-in time and gets the correct 3:00 PM answer. But if the guest follows up with, "Great, and is my specific room ready now?" the chatbot fails. It has no connection to the live data in the PMS, so it cannot check reservation status or housekeeping updates. The guest gets a frustrating partial answer, and an employee still has to intervene.

Here is a common scenario. A guest texts the hotel two hours before arrival: "We're running a bit late, can we check in around 5 PM? Also, where is the best place to park a large SUV?" The PMS tool cannot parse this multi-part, unstructured request. The generic chatbot doesn't know the property's parking situation. A front desk employee must stop their current task, find the reservation in the PMS, add a note about the late arrival, type out detailed parking instructions, and send the reply. This single interaction consumes 10 minutes of staff time.

The structural problem is that PMS platforms are databases of record, and generic chatbots are stateless FAQ machines. Neither is built to be a stateful conversational agent that can authenticate a guest against a reservation, query a live database, parse natural language, and execute tasks. Bolting them together creates disjointed workflows that still require constant human oversight, defeating the purpose of automation.

The first step is a discovery audit of your current guest communication workflow and PMS. Syntora would review your PMS API documentation (whether from Mews, Cloudbeds, or another provider) to map out available data points like reservation details, room status, and guest profiles. We would also analyze your past guest messages to identify the 25 most frequent questions. You receive a clear scope document that outlines exactly which inquiries the AI will handle and the precise integration points with your systems.

The technical core of the system would be a FastAPI service deployed on AWS Lambda. When a guest sends a message via SMS or web chat, it triggers the service. The service uses the Claude API to understand the guest's intent and extract key information. If a guest asks, "Can you recommend a quiet coffee shop nearby?" Claude parses the intent and entities. The FastAPI service then queries a Supabase database of curated local recommendations and formulates a helpful response. For reservation-specific questions, the service authenticates the guest and queries the PMS API in real time.

The delivered system is a backend AI engine that integrates with your existing guest-facing channels. Your team gets a simple dashboard built with Vercel and Supabase to monitor conversations and handle any escalations. The system would automate the predictable, repetitive questions, freeing your 15-person team from the message inbox. They can then dedicate their full attention to providing exceptional in-person service, which is what truly defines the guest experience.