A booking bot for a restaurant: what it can do

A booking bot for a restaurant answers guests in a messenger and takes them from the question “is there a table on Friday?” to a finished booking: it checks free places, books the guest in and, when needed, moves or cancels the booking. Anything that doesn't fit the script it passes to a member of staff. Below we look at what exactly can be trusted to a bot, where the line runs, and how to organise the work so that it helps rather than creates new problems.

What the bot does

A good booking bot doesn't chat; it carries out specific actions. There aren't many of them, but they cover most ordinary requests.

• It checks free times and places. A guest writes: “Four of us, Saturday, around eight.” The bot looks at the real availability rather than guessing.
• It shows the list of tables. If the venue has a main hall, a terrace and a VIP room, the guest can choose where to sit.
• It creates the booking. The bot always finds out three things: the name, the phone number, and the place and time. Without them there is no booking.
• It finds the guest's bookings. “What do I have booked for the weekend?”
• It moves a booking. “Could we make it an hour later?”
• It cancels a booking. The guest wrote that their plans changed, and the table is freed for others.
• It answers questions about the venue. The menu, terms, dress code, age restrictions, offers, cancellation rules. But only what the owner has entered in the venue profile and uploaded in the menu beforehand.
• It calls a person. If a question is beyond its limits, the bot hands over the conversation, and the staff get a notification.

A few more properties that matter more in practice than they seem to.

It answers in the guest's language. The guest writes in Russian, English, Romanian or Armenian, and the bot answers in the same language. For venues visited by both tourists and locals, this settles the question “who on the night shift speaks English?”.

It understands voice messages. The bot turns a guest's voice message into text. For this, a speech recognition service has to be connected on the platform side.

It works only with the guest's own bookings. Ownership is checked by a confirmed phone number. In Telegram the bot asks the guest to press the “share phone number” button, so as not to rely on what the guest typed by hand. Someone else's booking can't be moved or cancelled through the bot.

What the bot doesn't do

An honest list of limits is more useful than a long list of capabilities. If you're setting your staff's expectations, start with it.

• It doesn't take payment. The bot can only tell the guest the deposit or minimum spend terms as text, the way the owner wrote them. It doesn't take money and doesn't calculate anything.
• It doesn't make up rules. If the venue profile has no answer, the bot doesn't improvise; it calls a person. More on preparing the materials: how to set up a booking bot.
• It doesn't confirm a booking it didn't create. There's protection against false confirmations: the bot won't say “you're booked” if there is no booking in the system.
• It doesn't settle disputes. A complaint, special terms for a group, a turnkey event: that's the manager's job.
• It doesn't replace the dashboard. Seating, the live floor and the “didn't come” mark are handled by the staff.

In this sense the bot isn't the manager's competitor but the first line: it takes away the routine and leaves people what needs decisions. We discuss this at length in the article AI or human front desk.

How a booking is taken

Take an example. A guest writes at 23:40, when the venue is closed.

1. “Good evening, is there a table for six on Saturday?” The bot asks for the time.
2. The guest replies: “At 20:00.” The bot checks the tables and offers options, for example the main hall and the terrace.
3. The guest chooses the terrace. The bot asks for a name.
4. Then it asks for the phone number. In Telegram the bot asks the guest to confirm the number with the “share phone number” button.
5. The bot creates the booking and only after that says it's made.
6. In the morning the manager opens the dashboard and sees the booking in Saturday's list, together with the conversation.

Now the key point: a double booking of one table for overlapping times is impossible. This is checked in the database, so if the table is already taken at 20:00, a second booking for it won't be created.

The bot doesn't replace “hard” rules; it keeps them. If you want to understand how such rules work in general, read the article about double bookings.

When the bot calls a person

Handing a conversation to a person isn't a failure but a regular part of the work. The bot calls a manager when:

• the question goes beyond what it knows (the venue profile, the menu, the ready answers);
• the guest asks for something out of the ordinary;
• a decision is needed that the bot shouldn't make.

At that moment the staff get a notification in Telegram. Then there are two ways: a member of staff answers at once, or comes back to the question when they're free.

An example of wording that helps the guest not to feel abandoned: “I've passed this question to the manager, who will answer you right here.” What matters is that the message is honest: don't promise exact times if you're not sure.

How a member of staff takes over a chat

Often you don't need to “call for help” but to step in yourself. The dashboard's Conversations section holds all the bot's conversations with guests, with search and filters. A member of staff opens the conversation and answers from the dashboard, and the bot goes quiet in that conversation while a person is handling it. This prevents the awkward situation where the manager and the bot write to the guest at the same time.

If the member of staff hasn't written anything for 12 hours, the bot returns to the conversation. This rule keeps a conversation from “hanging” unanswered after a person got distracted or the shift ended.

A practical tip: decide in advance who on your shift is responsible for conversations. If the notification went to everyone and nobody took it, the situation is worse than without a bot. The simplest way is to assign it to one role, for example the manager on duty.

What the owner sees and how to keep control

Automation without control is frightening, and rightly so. What the dashboard has:

• Conversations. You can read exactly what the bot answered the guest.
• The log. Who did what: a booking created, moved, cancelled.
• Permissions by role. Not every member of staff should see everything.
• Limits. The bot is protected against spam, and AI spending is capped so that stray traffic doesn't get out of hand.

The venue's data is isolated from the data of other venues. A venue can also forbid the use of its conversations (anonymised) to improve the service.

Where to start testing

Before letting the bot loose on real guests, go through the scenarios yourself. Write to it as a guest: ask for a table for tomorrow, move the booking, cancel it, ask about the menu and about something that isn't in the venue profile. The last question is especially telling: the right reaction to something unknown is handing the conversation to a person, not a confident made-up answer. Every case where the answer didn't satisfy you tells you what to add to the venue profile or the ready answers.

What it looks like in GuestKeepr

The GuestKeepr virtual front desk works in Telegram and WhatsApp. You connect the channels, set up tables and zones, fill in the venue profile, upload the menu (PDF, photos, Excel, Word, text) and invite your staff. We wrote in detail about how not to go wrong with channels in the article about bookings via WhatsApp and Telegram.

If a member of staff cancels or moves a booking in the dashboard, the guest gets a message in their messenger. This works for those who have written to the bot; in WhatsApp the 24-hour message window applies. A booking whose guests are already seated can't be cancelled or moved.

Frequently asked questions