Guests message at 23:40 asking whether breakfast is included, whether they can check in early, and how far the hotel is from the station. Nobody answers until 8:00. By then, one has booked elsewhere and another has checked in without ever hearing about the airport transfer. A hotel AI concierge exists to close that gap: it answers pre-booking questions instantly, handles routine guest messaging automation after booking, and offers the extras that lift revenue per room. This article covers what guests actually ask, where AI helps and where it does not, how upsells work, and how direct bookings shift the economics.
What hotel guests ask an AI concierge before booking
Pre-booking questions cluster around four themes: location and transport, rooms and amenities, price and cancellation terms, and availability for specific dates. "Is parking available?", "Do you have a room with two beds for the 14th?", "Can I cancel free of charge?", "Is the area quiet at night?" These are not complex questions, but they arrive at inconvenient times and they decide whether a booking happens at all.
The timing problem is well documented. According to Drift research covering 30 million conversations, 39% of website chats start outside working hours. For a small hotel with no night receptionist, that is a large share of demand hitting voicemail. HubSpot Research also found that 82% of buyers consider an immediate response important, where "immediate" means ten minutes or less. A guest comparing three boutique hotels on a Friday evening will book with whoever answers first.
A well-configured hotel AI chatbot answers from a knowledge base the hotel controls: room types, policies, distances, check-in times, parking, pets, accessibility. It does not need to be creative. It needs to be accurate, fast, and able to hand over to a human when the question falls outside its scope.
Where AI helps and where it does not
What it handles well
Repetitive, factual questions are the natural territory: check-in and check-out times, Wi-Fi, breakfast hours, cancellation policy, directions, parking cost, pet rules, and the status of an existing reservation. It also handles multilingual replies, which matters for city hotels with mixed international demand. And it works at 2:00, which is when a meaningful share of pre-booking research happens.
What it should not do
An AI concierge should not improvise on price negotiations, complaints, refunds, safety incidents, or anything involving a guest's personal data beyond the reservation record. It should not promise an upgrade that the front desk cannot deliver. The practical approach is a clear escalation rule: if the intent is not in the knowledge base, or the guest expresses frustration, the conversation goes to a human with the full transcript. Our guide to guardrails for clinics and hotels covers the policy layer in more detail.
There is also a commercial limit. AI improves response speed and coverage; it does not fix a weak rate strategy or a poor location. Treat it as a conversion and service layer, not a revenue strategy on its own.
Upsells: breakfast, late checkout, transfers
Upsell revenue is usually lost through omission rather than rejection. Guests do not ask about late checkout because they do not know it is possible; they do not book the transfer because nobody offered it. An AI concierge can offer these at the right moment in the conversation, using the same rules the front desk would apply.
Three upsells work well because they are easy to describe and easy to price:
- Breakfast. Offered at booking or the evening before arrival, with price per person and dietary options. Conversion is highest when the guest is already thinking about the morning.
- Late checkout. Offered the evening before departure, subject to housekeeping availability. This is a capacity decision, so the AI should check a rule or a simple availability flag before confirming.
- Transfers and parking. Offered when the guest asks about arrival, which is one of the most common pre-arrival questions. A fixed price and a clear meeting point remove friction.
Each offer should be a single, clear question with a price attached, not a menu. "Would you like to add breakfast for two at 18 per person?" outperforms a list of five extras. Every accepted upsell should post to the PMS or be logged for the front desk, otherwise the guest arrives to a surprise.
Direct bookings versus OTA commission
Every booking that arrives through an OTA carries a commission. Booking.com states that commission is charged on every booking and the percentage depends on country, property type and contract; it does not publish a single global rate. Skift's widely cited industry benchmark is around 15%, which is a benchmark rather than a guaranteed rate.
The arithmetic is straightforward. Take a 30-room hotel with an average daily rate of 200 GEL and 60% occupancy. That is 18 rooms sold per night, or roughly 540 room-nights a month. At 15% commission on 200 GEL, each OTA booking costs about 30 GEL. If 40% of those room-nights come through OTAs, the monthly commission bill is roughly 6,480 GEL — an estimate based on those assumptions, not a measured figure.
A hotel AI concierge does not replace the OTAs, and it should not try to. What it does is capture the guest who is already on the hotel's own site or messaging the hotel directly, answer them before they return to the OTA, and offer a direct rate or a small incentive such as flexible cancellation or breakfast. Over a year, shifting even a modest share of bookings from OTA to direct changes the margin materially. For a fuller comparison of channel economics, see direct bookings versus Booking.com.
How to measure the impact
Measure four things, and measure them before and after launch.
Conversion and response metrics
Track the share of website conversations that end in a booking or a booking link click, the median first-response time, and the share of conversations outside working hours that receive a useful answer. A reasonable target is a first response under one minute, with escalation to a human within ten minutes during staffed hours.
Revenue metrics
Track upsell attach rate per booking, upsell revenue per room-night, and direct booking share month over month. Attribute upsells to the conversation that produced them; without attribution, the front desk and the AI will both claim credit.
Cost and quality metrics
Track the share of conversations escalated to staff, the number of factual errors reported by guests, and the hours of front-desk time saved on repetitive questions. That last figure is an estimate unless you time it, so sample a week of transcripts rather than guessing. On cost, see what an AI concierge costs for a realistic range.
See it in action
The hotel demo shows the breakfast and late-checkout questions, the transfer offer and the booking flow end to end, using the same 30-room example above.
What to do next
- Export the last 90 days of guest messages and website chats, and group them into ten intents. That list becomes your knowledge base and your escalation rules.
- Define the three upsells you will offer, with fixed prices and availability rules, and decide how each one posts to the PMS.
- Set a baseline for response time, direct booking share and upsell revenue before launch, so the comparison is honest.
- Shortlist two or three vendors and test them on your own ten intents; our guide to how to choose a vendor lists the questions to ask.