AI customer support for restaurants

The dinner rush and the message rush land at the same time, and the phone is usually already in someone's hand taking an order. This page is not the generic case for 24/7 AI support. It is what an AI agent actually does for a restaurant: table reservations, menu and allergen questions, catering enquiries, a photo of a wrong order, and tonight's special reaching the regulars who asked to hear about it.

The rush doesn't wait for someone free to check WhatsApp

A restaurant's messages don't arrive evenly through the day. They bunch in the ninety minutes before service, exactly when the same two or three people who might answer a DM are seating walk-ins, running food and taking phone orders. A question that would take ten seconds to answer at 3pm sits unread at 7:15pm, and by the time someone gets to it the customer has usually already booked somewhere else or called a competitor.

An AI agent doesn't need a quiet moment. It reads the menu question, the reservation request or the "are you open tonight" message the second it lands, answers from what you've told it, and only pulls a person in when the answer genuinely needs one. The busiest hour of your day is the one this actually matters for, not the slow Tuesday afternoon.

Reservations without a phone that's already busy

Replio's booking calendar is free on every plan and takes appointments straight inside the chat on WhatsApp or Telegram, not just from a link. For a restaurant that means setting up event types the way you'd actually describe seatings, such as "Table for 2", "Table for 6 or more" or "Sunday brunch", each with its own duration, a buffer between seatings, and a notice window so someone can't book a table for twenty minutes from now during Friday dinner.

A customer gets a reminder before their booking, a self-serve link to reschedule or cancel, and a downloadable calendar invite. You also get a public booking page at replio.live/book/ to put in your Instagram bio and on your Google profile, so someone who finds you there can book a table without waiting for anyone to reply. After the sitting, staff mark it completed or no-show, which is the honest way to see which nights are actually losing covers to people who never show up.

Where this stops: the calendar doesn't see your floor plan or your point-of-sale system. If a night is fully booked on paper but a table opens from a cancellation, someone still has to update availability in the Replio dashboard. It replaces the phone call and the paper diary, not table management.

The menu is the knowledge base, and it has to stay one

Upload your menu, and a customer can ask "is the laksa gluten-free", "what's in the Sunday roast" or "do you have anything vegan" and get an answer pulled from what you actually serve, not a generic response about dietary options in general. The same knowledge base holds today's price list, so a price quoted on WhatsApp matches what's on the till.

Say this plainly, because it matters for a kitchen: the AI only knows what you've told it. If a recipe changes, an ingredient gets substituted, or a dish is 86'd for the night, that has to be updated in the knowledge base the same day, or the AI will confidently repeat the old answer. For a severe allergy, treat an AI answer as a starting point and have the customer confirm with a staff member before the food leaves the kitchen, the same caution you'd want from a printed menu with an asterisk on it.

Catering and private-event enquiries don't wait for opening hours

A message asking about a 40-person birthday dinner or an office holiday party rarely arrives at 2pm on a Tuesday. It arrives at 11pm on a Sunday, from someone planning three months ahead, and it sits on somebody's personal phone until they get around to it, if they remember at all. Replio's lead capture asks the qualifying questions in chat, the date, headcount, occasion and rough budget, logs the enquiry, and hands a fully-formed lead to a person the next morning instead of a half-remembered DM.

A photo says more than a paragraph

Replio reads images, so a customer can send a photo of the wrong dish, a cold delivery, or a receipt they have a question about, instead of typing out a description that loses detail in translation. The complaint gets logged and, if it needs a person, escalated into the shared inbox during that same shift, not left for whoever opens the account the next morning to discover a one-star review has already gone up in the meantime.

One inbox for a dining room that doesn't share a language

A single evening's covers regularly span several languages, and the AI detects and replies in each customer's own language automatically, across 50+ languages, from one knowledge base rather than a separate one per language. It also transcribes voice notes, so a customer who'd rather ask "does the curry have shellfish in it" out loud gets answered the same as someone who typed it. On Scale and Pro, Live Translate lets your own team read and reply to any language inside the shared inbox too, which matters on the nights a booking needs a human and nobody on shift speaks the customer's language.

Instagram, where the food photos live

Restaurants post food, and a photo of tonight's special draws the same questions every time: "price?", "do you deliver?", "is there a vegan version?", "can I book for 8?". Most of them arrive as Instagram DMs, which is why it's the channel restaurant owners usually ask about first. Where Instagram stands today: answering Instagram DMs for every customer needs a permission from Meta, instagram_manage_messages, which Meta has not yet approved for Replio. You can connect your Instagram account now, and replies start for everyone once it is approved, with nothing to change on your side. Automatic replies to comments under your posts depend on the same kind of Meta approval, so don't plan a launch around them yet.

What works today is sending a photo's audience somewhere that answers. Put your WhatsApp chat link and your replio.live/book/ booking page in the Instagram bio, and write "book via the link in bio" in the caption of the posts people actually respond to, rather than "DM us". WhatsApp, Messenger, Telegram and the website widget are live for every customer today, so the person who asks "gf?" at 11pm gets an answer from the current menu instead of a DM nobody reads until tomorrow.

Tonight's specials, sent to people who actually opted in

WhatsApp broadcasts send an approved template to contacts who opted in, handle STOP opt-outs automatically, and the AI answers whatever replies come back, "still doing the tasting menu tonight?", "what time do you close?", "can I still book for 8". A broadcast that generates ten replies you have to answer by hand isn't actually saving anyone time; one that generates ten replies the AI answers on its own is.

A Friday night, one hour at a time

6:15pm: a WhatsApp message asks for a table for four at 8, and Replio's calendar confirms it and sends a reminder for two hours out. 8:40pm, mid-service: a WhatsApp photo shows a plate with the wrong sauce; the AI logs it as a complaint and pings the floor manager in the shared inbox, who comps the dish before the table finishes the rest of theirs. 9:50pm: a message asks about hosting a 30-person work party in November; the AI asks headcount and date, then hands a qualified lead to the events person for Monday morning. 11:05pm, after close: someone who saw tonight's dessert on Instagram taps the WhatsApp link in the bio, asks "gf?", and gets the allergen note pulled straight from the current menu. None of that needed a person free at the exact moment it happened.

Setting it up

1. Connect the channels your customers already use. Start with WhatsApp, where most diners already message you, and add Messenger if your Facebook Page gets questions; a takeaway-only spot might start with just one plus the free website widget. Connect Instagram too if you like: its DMs start flowing once Meta approves access.
2. Upload your menu, allergen notes and current prices as the knowledge base, and set a habit for updating it the same day anything changes.
3. Set up the booking calendar with event types that match how you actually seat people, plus buffers and a notice window.
4. Put your booking page and WhatsApp link in your Instagram bio and in the captions of your food posts, so the people a photo reaches land on a channel that answers.
5. Invite your team to the shared inbox so a complaint photo or a catering lead reaches a person the moment the AI escalates it.

Pricing for a restaurant

Starter at $49/month covers one channel (WhatsApp, Instagram or Messenger, your pick) plus Telegram and the website widget, unlimited customer messages and 2 seats, which fits a takeaway or delivery-only kitchen that lives mostly on one channel. A sit-down restaurant that takes bookings on WhatsApp and questions on its Facebook Page usually wants Scale at $149/month, which covers WhatsApp, Instagram and Messenger together plus Telegram and the widget, 5 seats and Live Translate included free. Pro at $349/month adds every channel, 15 seats and a branded Telegram bot, which suits a multi-location group. AI replies are 8,000, 30,000 and 50,000 a month on the three plans. The booking calendar, image reading and broadcasts are included, with no separate charge for any of them, and every paid plan starts with a 7-day free trial.

Frequently asked questions

Related