Saturday night, full dining room, two servers on the floor. The phone rings for the fourth time in ten minutes: someone wants to move a booking, someone else is asking whether you have gluten-free options. You pick up on the third ring — or you don't, and that booking goes to the restaurant next door.
It is the most ordinary and most expensive problem in hospitality: requests arrive exactly when you cannot handle them. An automated assistant on WhatsApp is not there to replace anyone — it is there to filter the hours when nobody is free to answer.
What happens today, in numbers
Measure before you automate. In a mid-sized restaurant, inbound calls cluster in two windows: 11:00–12:30 and 18:00–20:00 — precisely when the kitchen starts and the dining room fills up. Over a typical week the arithmetic looks like this:
- a booking call takes between 90 seconds and 3 minutes, including small talk, checking availability and confirming;
- a sizeable share of calls are not bookings at all but repeat questions: opening hours, parking, the menu, whether there are coeliac options;
- every call missed at peak time is, more likely than not, a table booked somewhere else.
The cost is not only the lost table — it is attention. Whoever answers the phone mid-service stops serving, and service quality drops at exactly the busiest moment.
What an assistant can handle, and what it cannot
The practical rule: automate whatever is repetitive and verifiable, and leave people everything that needs a decision or a relationship.
Good candidates
- Capturing the booking: date, time, number of covers, name and phone number.
- Allergies and intolerances: captured in writing, so nothing is misheard — this is where the phone fails most often.
- Takeaway orders: what, when, who collects.
- Frequent questions: hours, closing days, parking, whether dogs are welcome, whether there is a children's menu.
- Confirmations and reminders the day before — the single most effective lever against no-shows.
Leave these to people
- Large groups, private events, agreed menus: these are negotiations, not forms.
- Complaints. An unhappy customer is always handled by a person.
- Overbooking and exceptions: whoever knows the room decides whether tables can be squeezed.
A well-configured assistant does not try to close these conversations. It recognises them and hands them to the staff with the context already collected, so whoever picks them up is not starting from zero.
How a conversation should be structured
This is the whole difference between an auto-responder and a useful assistant. An auto-responder sends a message and waits. An assistant leads: it works out what the customer wants, asks only for what is missing, and closes with a complete record.
In practice, from "I'd like to book for Saturday" the assistant has to reach a record with date, time, covers, name, phone and notes — asking two or three questions at most. Ask eight, and the customer gives up and calls instead.
What reaches the floor is not a chat to read: it is a line ready to go into the booking book. You can see a full example of this conversation, with the fields filling in as the customer types, on the restaurants use-case page.
What you need to start
- A WhatsApp Business number. If you already use the WhatsApp Business app on the venue's phone, the number has to be migrated: automation requires API access, which is a different thing from the app. We covered it in the WhatsApp Business API guide.
- Your rules: service hours, closing days, capacity, how far ahead you take bookings, what to do when you are full.
- One destination for the data: the booking book you already use, a shared sheet, or the assistant's dashboard. What matters is that there is only one — two parallel lists quickly become two wrong lists.
- A named person who receives the escalated conversations. They do not need to sit in front of a screen; the notifications just need to land where they will see them.
The mistakes we see most often
Automating everything on day one. Start with bookings and the three most frequent questions. Add the rest once you have seen two weeks of real conversations: customers ask things you never anticipated.
Hiding that it is an assistant. Saying so works better: people are more direct, skip the pleasantries, and the conversation closes sooner. Pretending to be a person only produces distrust once the pretence breaks.
Not planning the exit. There must always be an obvious way to reach a human. An assistant that cannot say "let me pass you to the floor" is a wall, and walls lose customers.
In short
A WhatsApp assistant does not add covers by itself: it stops losing the requests you drop today, and frees people during the hours they are needed on the floor. It is an intervention on the process, not on marketing — which is why it is easy to measure. Count the bookings captured outside opening hours in the first month and compare them with the calls you used to miss.