WhatsApp chatbot examples, and how to plan your own

Read a week of WhatsApp chats and count what people actually ask. For most businesses a handful of questions cover a large share of the messages: where’s my order, when are you open, how do I return this, can I talk to someone.

Those are your menu. Anything rare goes to a person.

Every option should end in one of three ways: an answer, a few questions and then a hand-over with the answers attached, or a hand-over straight away. Write it down for each option before you build anything.

WhatsApp shows a few options as buttons and more as a list the customer opens. Tapping is faster than typing and leaves no room for spelling. Keep labels short enough to read at a glance; long ones get cut.

Still accept typing: people will reply “1”, “track” or the whole label, and the bot should understand all three.

When you need details, ask for them one by one, with an example of the format (“your order number, for example TH-104233”). If the answer doesn’t fit, say what you expected and ask again, then move on after a second miss rather than looping.

Save what you collect where your team will see it, so nobody asks the customer again after the hand-over.

Put “Talk to our team” in the main menu, and let the customer type words such as “agent” or “staff” at any point. If someone misses the menu a few times in a row, offer the person option rather than repeating the menu.

Outside your hours, say when you’re back and take a message instead (see the after-hours guide).

Short messages, one idea each, the customer’s name if you have it. Split a long answer into two or three messages with a short pause between them, the way a person types. Skip the jargon and the exclamation marks.

Run every path on a real phone before it goes live, in each language you support. Once it’s live, look each week at where people leave, which options lead to hand-overs, and what they typed that the bot didn’t understand. That list is your next round of changes.