Putting AI on your WhatsApp without losing the customer

For a lot of service businesses, WhatsApp is the front desk. Bookings, questions, complaints and reorders arrive in the same chat list, often after hours and often faster than staff can answer. It is the obvious place to put AI, and the easiest place to get it wrong.

Answer from your records, not from guesswork

A general-purpose model will happily invent a price, an opening hour or a table that does not exist. An assistant that speaks for your business has to answer from your own menu, prices, rules and availability, and say so plainly when it does not know.

Decide what it may do on its own

Answering a question about opening hours is low risk. Promising a refund, changing an order or handling a complaint is not. Set hard limits on what runs unsupervised, route everything else to a person, and measure how often the answers are right before you widen those limits.

Make the handover instant

Customers rarely mind talking to an assistant. They mind being stuck with one. Staff should be able to take over any conversation in a tap, with the whole history in front of them, and the AI should step back the moment they do.

Respect the channel

WhatsApp has rules for businesses, and they exist for customers. Honour every request to stop receiving messages, send marketing only to customers who expect to hear from you, and keep the conversation on the number your customers already know.

Keep the customer data the business's

The conversations, bookings and visit history belong to the business, not to the tool. Before switching anything on, check where the data is processed, whether it is used to train anyone's models, and how you would take it with you if you left.

Those are the decisions we built Meta-Koda around: an AI that answers on the WhatsApp number you already use, from your own records, and steps back whenever your staff step in.