A practical design method for support chatbots: choose intents, keep flows short, avoid dead ends and hand complex or sensitive requests to a person.
Choose intents from real conversations
Review recent support chats and group them by customer goal: track an order, understand a product, change an appointment, report a problem or reach the team. Use the customer’s wording when naming options. A menu with internal department names forces people to translate their need into your org chart.
Pick two or three intents with clear answers and low risk. Leave account disputes, safety concerns, complex refunds and anything requiring judgment for a trained agent.
Build a short path with a useful fallback
Ask only for information needed to route or resolve the request. Tell the customer why you need each detail. Use short steps, confirm important choices, and provide a way to go back or speak with a person.
A fallback should not loop the same prompt. If the bot cannot understand twice, offer a human handoff or a small set of clear alternatives. When the support team is offline, explain when it will respond and capture only the details needed for follow-up.
- State what the bot can help with.
- Offer a few common customer goals.
- Collect the minimum context for the chosen goal.
- Confirm the action or explain what happens next.
- Transfer unresolved requests with the chat context attached.
Keep answers grounded in current business information
Store the approved answer source for policies, opening hours, delivery areas and product details. Assign an owner to review those facts when the business changes. Do not let a chatbot invent a refund promise or availability when the source is missing.
If AI is used to phrase answers, constrain it to reviewed business information and test ambiguous questions. Give agents a way to see what the customer asked and what the automated system already told them.
Test the exceptions, not only the happy path
Test typos, mixed-language requests, an abandoned conversation, a customer who changes their mind and an out-of-scope question. Ask someone unfamiliar with the flow to try it without coaching. Record where they hesitate, then simplify the wording.
After launch, review failed handoffs, repeat questions and customer feedback. Add a new intent only when you have a reliable answer and a responsible team to own it.
Common questions
Should a support chatbot answer every WhatsApp message?
No. Keep automation to predictable tasks and offer clear escalation for unusual, account-specific or sensitive issues.
What should happen when the chatbot cannot help?
Offer a direct human handoff or collect a minimal summary and tell the customer when the team will respond.
References
Platform features and policies can change. Confirm current requirements in official documentation before implementation.