Building a WhatsApp Chatbot That Customers Don't Hate

A good WhatsApp bot answers in one tap and hands over fast. A bad one traps people in menus. The difference is design, not technology.

Start with the ten intents

Pull 500 recent conversations and tag them. In almost every business, ten intents cover the bulk of volume:

1. Where is my order
2. Pricing / plans
3. Product availability
4. Return or refund status
5. Booking or rescheduling
6. Payment failed
7. Business hours and location
8. Talk to a human
9. Documents / invoice copy
10. Complaint escalation

Automate these ten completely. Everything else goes to an agent.

Menu bot vs AI bot

Menu (button) bots are predictable, cheap and fast — best for transactional intents where the answer depends on data lookup. AI bots handle free text and long-tail questions well but need guardrails and a knowledge base. The strongest pattern is a hybrid: buttons for the top ten, AI for anything typed, human for anything unresolved in two turns.

Design rules that matter

• Three options maximum per message. WhatsApp allows three quick replies; more becomes a list nobody reads.
• Always show an exit. Include "Talk to a human" from the first message.
• Confirm before acting. Cancellations, reschedules and refunds need an explicit yes.
• Use typing indicators and 1–2 second delays so replies feel conversational instead of robotic.
• Never loop. If the bot fails twice, escalate automatically and say so.

Handoff done right

When a bot escalates, pass the full transcript, the detected intent and customer data (order ID, plan, ticket history) into the agent's inbox. Set expectations in the message: "A specialist will reply within 10 minutes (Mon–Sat, 10am–7pm)."

Measuring the bot

Track containment rate (resolved without a human), escalation rate, average turns to resolution, and CSAT after bot-only conversations. A containment rate above half with stable CSAT means you can safely automate the next tier of intents.

Build for the boring questions first. The revenue effect comes from freeing agents for the conversations that actually close deals.