Building a WhatsApp chatbot that customers actually like
A practical guide to WhatsApp chatbots and automation — when to automate, when to hand off to a human, and the flows that qualify leads without annoying people.
A good WhatsApp chatbot is invisible — customers feel helped, not processed. A bad one is the reason people reply 'human' within ten seconds. The difference isn't the technology; it's how thoughtfully you design the conversation. This guide covers the patterns that work and the mistakes that drive customers away.
What to automate (and what not to)
Automate the repetitive, low-judgement parts: answering the same five FAQs, capturing lead details, confirming bookings, giving order status, and routing conversations to the right agent. Keep humans for the high-stakes parts: complaints, complex queries, anything involving money, and any conversation where a customer sounds frustrated.
The best bots make the hand-off to a human seamless. If your bot can't say 'I'm passing you to someone who can help' and actually do it within seconds, it's not automation — it's a wall.
The welcome-bot pattern
The most effective automation for new conversations is a simple welcome bot triggered by keywords like 'hi' or 'hello'. It greets the customer, sets expectations ('I'm a bot — I can help with X, Y, Z, or connect you to a person'), and offers a small menu of options. Each choice routes the conversation: answer an FAQ, capture details onto the contact, or escalate to an agent.
Crucially, replies inside the 24-hour customer-initiated window are free. So an entire welcome-bot conversation — qualifying a lead, answering questions, capturing a phone number — costs you nothing in message fees. This is the single best use of automation.
Capturing details with parameters
Every answer a customer gives can be saved onto their contact record as an attribute — name, city, budget, product interest, preferred time to call. This turns a chat into structured data you can segment and follow up with later. A welcome bot that asks 'which city are you in?' and stores the answer lets you later run a city-specific campaign to everyone who answered 'Mumbai'.
Plan the attributes you want before you build the bot. The most valuable for SMBs: name, city, product interest, budget range, and whether they've bought before. Don't over-collect — every extra question is a drop-off point.
Honouring STOP automatically
When a customer replies STOP, UNSUBSCRIBE, or STOP ALL, WhatsApp expects you to opt them out of marketing immediately — regardless of what your bot or flow is doing. Your platform should enforce this at the system level so no flow can override it. This isn't optional; it's compliance, and it protects your number from being banned.
Good automation respects consent at every step. A bot that keeps messaging someone who opted out is a bot that gets the whole account suspended.
Measure what matters
Track three numbers for every bot: completion rate (what share of conversations reach the intended end), hand-off rate (what share escalate to a human), and CSAT after resolution. If your hand-off rate is above 40%, your bot is doing too much — push more resolution into self-service. If it's below 5%, you might be hiding human support from people who need it.
The goal isn't to replace humans. It's to free humans for the conversations where they add real value, and let the bot handle the rest instantly, any time of day.
Create a free AiConva workspace and put these ideas into practice in minutes.