How to build a WhatsApp chatbot for your restaurant

Start from what customers actually send. Guests ask whether there is a table tonight, whether the kitchen can do a dish without nuts or dairy, how late you serve, whether there is parking, and whether a large group can have a set menu.

Then look at what goes wrong today:

• No-shows burn your revenue daily. Empty tables on Friday evening — booked, never showed, no call. 4-5 tables a week = pure lost revenue.
• Guests come once and are never heard from again. A great evening ends at the door. Nobody has their number, nobody follows up, and the table sits empty next Friday.
• Instagram DMs sit unanswered for hours. "Table for 2 tonight?" asked at 6 PM. You reply at 9 PM. They've already eaten elsewhere.

Each of these is a conversation with a predictable shape — a question, a lookup, a confirmation. That shape is what a chatbot handles well.

1. How many guests? Buttons for 2, 4 and "more" — larger groups go down a separate path.
2. Which day? Today, tomorrow, or pick a date.
3. What time? A list message of the open slots for that party size, not every half hour of the evening.
4. Any preference or occasion? Indoor or outdoor, high chair, birthday — optional, one tap to skip.
5. Name and confirm — returning guests are recognised by number and skip this.

If the chosen time is full, the bot offers the nearest two slots either side and the waitlist, rather than simply saying no.

Notice what the business side does: it answers with the specific detail (time, reference, next step), it asks for one thing at a time, and it gives the customer a one-word way to reply. Those three habits matter more than how clever the bot is.

"Can you send the menu?" is one of the most common messages a restaurant receives. The bot should answer instantly with a readable menu — a PDF or image that works on a phone screen, or a WhatsApp catalogue for dishes with photos and prices — plus a line on today's specials if someone updates them. Keep the menu current: an old menu with old prices causes more trouble at the table than no menu at all.

Reputoo supports both, and most restaurants end up using both:

• Keyword flows — a fixed menu or trigger word ("MENU", "BOOK", "STATUS") leads down a set path. Predictable, easy to audit, right for anything with a fixed answer.
• AI chatbot — reads the customer's own words and answers from your knowledge base (website, price list, policies). Right for the long tail of questions nobody wrote a menu item for.

A sensible start for a restaurant: keyword flows for the top two or three tasks, AI for everything else, and a person behind both.

On a full Saturday the bot can put walk-ins and callers on a waitlist with their party size and phone number, tell them roughly where they stand, and message them when a table is ready with [Coming] [Can't make it] buttons. A guest who does not reply within a few minutes loses the table to the next on the list. The host stand gets a queue that manages itself, and guests can wait in a nearby café instead of the doorway.

"Is the biryani gluten-free?" or "do you have Jain options?" can be answered by the bot from a maintained menu with dietary tags. "My son has a severe nut allergy — is it safe to eat there?" cannot, and the bot should say plainly that the manager will reply, and pass the chat on. Get this wrong once and no automation saving is worth it.

• Complaints, refunds and anything where the customer is upset.
• Anything involving money the bot cannot see — a disputed bill, a custom quote.
• Health, safety or legal questions.
• Any customer who asks for a human. Do not make them ask twice.

For restaurants specifically: allergies, large-party deposits, complaints about a meal, and anything about a private event should reach the manager, not a script.

Handover on Reputoo moves the chat into the shared inbox with the conversation so far, so staff never ask the customer to repeat themselves.

1. Connect your WhatsApp number to Reputoo (official API; takes minutes).
2. Add your website and documents to the knowledge base so the AI answers in your restaurant's own terms.
3. Build keyword flows for the top tasks above. What to connect: most restaurants already run a POS and often a table-booking system; the useful connection is the guest record, so a returning diner is recognised.
4. Choose who gets handed-over chats, and during which hours.
5. Test from a personal phone, then share the number or a QR code with customers.

A booking for fifteen, a birthday dinner, a corporate lunch: these need a set menu, a deposit, maybe a private room. The bot's job is to collect the date, headcount, budget per head if offered, the occasion and any special needs, then hand to the events manager. The guest should get a clear "our events team will reply within the hour with options" rather than a generic "thank you".

For restaurants that take direct takeaway orders, the bot can show menu sections as list messages, add items, confirm the total and a pickup time, and send the order to the kitchen. Keep modifiers simple — spice level, extra portion, no onion — and pass anything complicated to staff. A bot that argues with a customer about a half-and-half pizza loses the order.

Guests write in the language they speak at the table and often mix: "table for 4 aaj raat 8 baje milega?". Keyword flows need those phrases; the AI chatbot understands the rest and replies in the guest's language. Dish names are the thing to watch — add the local names guests use for your dishes to the menu so questions about them are recognised.

• Book for two, for six and for fifteen, and check each takes the right path.
• Ask for a fully booked time and check alternatives and the waitlist are offered.
• Ask for the menu, and ask a dietary question the menu answers and one it does not.
• Cancel from the reminder and check the table is released.
• Place a takeaway order with a modifier and check the kitchen sees it.

Watch bookings completed without staff, cancellations made through the reminder rather than no-shows, and the phone during service — the calls that interrupt the host stand should fall first. Read the handovers weekly; a recurring question the bot cannot answer is usually a line missing from the menu or the FAQ.

Write the rules once and let the bot apply them the same way every time: how long a table is held for late guests, how much notice a cancellation needs, and what happens to a deposit. A guest running late can message "10 minutes late" and the bot confirms whether the table is still held. A guest who cancels in two taps is offered another night straight away. Exceptions go to the manager.

A handover should arrive with the booking or order, the guest's own words and the reason — allergy, large group, complaint, or a question the bot could not answer. The bot tells the guest who will reply and when, especially during service when staff are stretched. The host picks up a conversation that is ready to resolve, not one that starts with "which booking is this?".

The bot should know which days the restaurant is closed, when last orders are, and which nights run a set menu only — New Year's Eve, Valentine's Day, festival nights. Update these in one place before each special night, and the bot offers only what is actually available instead of promising à la carte on a set-menu evening.

A short message the day after a booking — "Thanks for dining with us. How was everything?" with [Loved it] [Could be better] — catches a bad evening while the guest still cares enough to tell you. "Could be better" goes to the manager with the booking attached. Send it once, keep it to one question, and never attach an offer to it.