What this automation does
Front desks and reservation teams answer the same twenty questions all day: what time is check-in, where do I park, can I check in early, is breakfast included, can you add a crib. This agent answers those with the reservation in front of it and passes everything else to staff with context.
- Recognize the guest. Inbound messages from SMS, WhatsApp or email are matched to a reservation by phone number, email or confirmation code.
- Send pre-arrival messages. A few days before arrival, the agent sends check-in instructions, parking details and an invitation to share arrival time or special requests. Restaurants send a confirmation with party size and a reply-to-cancel option.
- Answer questions from the property knowledge base. Check-in and checkout times, parking, pets, pool hours, breakfast, accessibility features, nearby transit. Answers come from a document your team maintains, not from the model’s general knowledge.
- Handle simple requests. Early check-in or late checkout requests are logged for the front desk to approve. Arrival times, crib requests or dietary notes are written to the reservation.
- Escalate the rest. Complaints, billing disputes, date changes, cancellations inside the penalty window and anything involving safety go to staff with a summary and the conversation history.
- Follow up after the stay. A thank-you message and a review request go out after checkout, and negative replies route to the manager before any public review prompt.
When it is worth building
This is worth it when your team spends hours a day in guest messages and many of those messages repeat. Independent hotels with a small front desk, rental managers with dozens of units and one guest-experience person, and restaurant groups whose host stands field constant text confirmations all fit.
It is not worth building if your PMS or a guest messaging product already handles this for your property type. Many PMS platforms include scheduled messages, and dedicated guest messaging products exist for hotels and rentals. A custom agent makes sense when you run several systems (a PMS, a restaurant platform and a spa booking tool, say), when you need it to act on reservations rather than just answer, or when off-the-shelf tools do not support your channel mix.
Architecture
| Layer | What we use | Why |
|---|---|---|
| Channels | Twilio for SMS, WhatsApp Business, Gmail or your reservations inbox | Where guests actually reply |
| Orchestration | n8n | Built-in Twilio and Gmail nodes, scheduled pre-arrival runs, simple handoff to staff |
| Reservation data | PMS or reservation platform API (for restaurants, OpenTable through its partner program), usually via partner or approved access | Lets the agent see dates, room type, party size and notes |
| Knowledge base | A structured property guide per location, kept in a doc or table | Staff update parking or pool hours without touching the agent |
| Reasoning and replies | Claude or OpenAI’s GPT models through the API | Understands guest questions in many languages and replies in your house style |
| Staff inbox | Slack, Teams or the PMS messaging inbox | Escalations arrive with the guest, reservation and conversation attached |
How to build it
- Pull a month of real guest messages. Group them by intent. Typically a handful of question types make up most volume. Those are what the agent should handle first.
- Write the property guide. One structured document per property with check-in rules, parking, amenities and hours, policies (pets, smoking, deposits, cancellations), and local directions. Mark which answers change seasonally.
- Confirm PMS access. Register for the partner or API program your PMS requires and decide read versus write scopes. This is often the longest lead-time item.
- Define what the agent may do. Write an explicit list: answer from the guide, log arrival times, add notes, request early check-in. Everything else escalates. Put these rules in code, not only in the prompt.
- Build the conversation loop. Match the guest, load the reservation, load the property guide, classify intent, then answer, act or escalate. Keep the last few messages as context.
- Add channel compliance. Register your SMS number for A2P 10DLC, record opt-in, honor STOP replies, and add a disclosure that responses are automated.
- Launch on one property in draft mode. For the first week, the agent drafts and staff send. Switch to auto-reply for the intents where drafts needed no edits.
What breaks in production
- Unmatched guests. Bookings made through an OTA may carry a relay email or no phone number. Ask for the confirmation code instead of guessing.
- Stale policies. The pool closes for maintenance and nobody updates the guide. Give staff a two-minute way to post a temporary notice the agent reads first.
- Group and multi-room bookings. One guest asking about “our rooms” may mean five reservations. The agent must recognize group blocks and escalate changes.
- Promises the property cannot keep. “Sure, early check-in at 10 a.m.” is not the agent’s call. Requests should always be logged as requests, with a clear “the front desk will confirm” reply.
- Language switching. Guests write in their own language mid-thread. Reply in the guest’s language, but keep internal summaries in English for staff.
- OTA messaging limits. Messages inside Booking.com, Expedia or Airbnb threads usually require certified connectivity. Plan for those guests to be handled by your PMS inbox until access is in place.
Timeline and cost
A single property with one PMS and SMS plus email channels typically takes three to six weeks, often longer if partner API approval is slow. Lumeor builds this in the range of $15,000 to $40,000, depending on the number of properties and systems, whether the agent writes changes back to reservations, and whether voice is included. We built messaging workflows on WhatsApp and Telegram in our n8n automations suite, and if you want phone reservations covered too, see how to build an AI voice agent for phone support. Our chatbot vs AI agent comparison explains why a reservation-aware agent costs more than a website chatbot.
How to measure success
- Share of guest messages answered with no staff involvement
- Median first-response time, day and night
- Front desk minutes spent on messaging per occupied room or per cover
- Early check-in and other paid requests captured
- Guest review mentions of communication and responsiveness