Skip to content

Blog · Business automation

Restaurants: which part of an order or a booking an agent can take over WhatsApp

Which AI WhatsApp agent can take orders and bookings for a restaurant in Colombia: tables, menu, delivery by neighborhood and payment in pesos.

Equipo editorial de Meteor Published Reviewed 4 min read

Which AI WhatsApp agent can take orders and bookings for a restaurant in Colombia?

One that checks the venue's system before committing to anything. An AI agent can take a booking against real table availability, build an order from the menu and prices that are loaded, answer where an order stands and say when something is unavailable. In Colombia, Meteor prices delivery with the fee you declare per neighborhood, collects in pesos through PSE or Nequi with Wompi and handles Rappi orders from the same place. What it cannot do is promise a table that does not exist, apply a price that is not on the menu or settle a complaint about the food: that goes back to a person at the venue, with the full conversation.

A full restaurant has a problem that never shows up at the till: the phone. It rings while somebody is taking an order, while a delivery app sends another one through, and while two people wait at the door. Almost all of those calls ask the same things: is there a table, what is on the menu, how much is it, where is my order.

Booking against the tables that exist

A booking confirmed over chat without checking real availability is a table somebody will have to invent at nine on a Friday night. The agent checks before confirming, exactly as the person holding the reservations book would, except it does not stop answering everything else.

The order, without double entry

When an order is taken over chat and then retyped into the system, two costs appear: the time and the mistakes. In the order-from-chat recipe the order lands directly in the POS, with the menu and prices the venue has loaded. With Fudo, the ticket is dine-in and tied to a table; with another POS, the order reaches your team in the conversation, already organized, to enter it once.

It is the least impressive part and the one that pays for itself fastest on a busy night.

Orders arriving through other apps

A venue with three channels does not have one problem per channel: it has the problem that nobody sees all three at once. Handling Rappi orders from the same place is not about saving tablets, it is about no longer having three separate queues competing for the same kitchen.

Restaurant orders and bookings in Colombia

Three things are specific to a venue in Colombia, and the restaurant solution covers them like this:

  • Delivery by neighborhood or by distance. The Met checks that the address falls inside the zones you serve and adds the fee you declare per neighborhood, which can be free above an order amount. If it is set up by distance, it charges by rings in kilometers from the nearest location, using the location the customer sends or their address. If the address falls outside, the order is flagged for your team to review. It does not connect to courier platforms or plan routes.
  • Payment in pesos. With Wompi it sends a payment link for PSE or Nequi and checks whether the transaction was approved. If the customer pays by bank transfer, it receives the receipt, and the order waits for your team to validate the payment before it goes to the kitchen. With cash or card on delivery, it can go straight through if you set it up that way.
  • Rappi. The integration is built on Rappi’s Colombian operation: the Met accepts the order with the real cooking time, marks it ready, switches off a sold-out product and closes the store when the kitchen is swamped. It does not change prices or edit the menu on Rappi.

Habeas Data in a restaurant

A WhatsApp order collects personal data: the name, the phone number and the delivery address. Colombia’s Law 1581 of 2012 applies just the same when an AI agent is the one asking:

  1. The first message informs. Who the restaurant is, what it will use the data for and where its data processing policy lives. If you later want to send promotions to that number, that purpose is stated from the start.
  2. Only what is needed. For a delivery, the name, phone number and address are enough.
  3. The card never goes through the chat. Payment happens on the payment gateway, and the Met does not store card details.
  4. A query or complaint about their data goes to a person. The law allows 10 business days for queries and 15 for complaints.

The restaurant is the data Controller; in Meteor, the platform acts as Processor, as clause 16 of its Terms sets out. The detail is in the guide “Habeas data and AI agents on WhatsApp: what Colombia’s Law 1581 requires”, on this same blog. None of this is legal advice.

What it must not touch

The menu rules. If a dish sells out and nobody marks it, the agent will keep offering it, because it reads what is there: that is not a flaw in the agent, it is a venue task made visible.

And complaints about the food go back to a person. The support Met hands over the full conversation so nobody has to ask again what happened.

Where to start

With the question that arrives most. In a restaurant that is usually “is there a table?” or “where is my order?”, and both have an exact answer in a system the venue already uses. The AI for restaurants that takes WhatsApp orders builds the rest on top of that: menu, delivery by neighborhood, hours, payment and order status.

If you want to hear it from a restaurant, in the Barril y Panceta case the team explains on video how it handles the delivery orders that reach it on WhatsApp today.

The limits

What this article does NOT answer

We would rather say it here than leave you hunting for something that is not there.

Keep going

What this article mentions, in detail

The integrations, the Mets and the automations named above, each with its own page.

Frequently asked

Questions on this topic

From the system where the restaurant already maintains them. If a dish sells out and somebody marks it, it stops being offered; if nobody marks it, the agent will keep offering it, because it reads what is there.

It helps by putting them in one place. The real problem of a venue with three channels is not each channel on its own, it is that nobody sees all three at the same time.

Yes: the Met takes the table's order and leaves it organized in the conversation so your team enters it once. With Fudo, the Met also creates the ticket on a table and it lands in the POS on its own; it is dine-in.

Keep reading

Other articles on the same thing

Want to see it running on your operation?

We show it with your own accounts connected, not with a canned demo.