WhatsApp & automation
Planning a WhatsApp chatbot with a reliable human handoff
Define answers, permissions and escalation before building a customer-facing bot.
Choose a narrow first job
Start with a repeatable task such as collecting an inquiry, answering approved FAQs or looking up order status. Write down the allowed questions, required data and conditions under which the bot must stop. A bot with a clear boundary is easier to test than an assistant told to handle everything.
Review representative customer conversations with personal details removed. Group the repeated questions and identify which answers depend on live systems. Do not train a response around an old spreadsheet if prices or delivery availability change frequently.
Separate answers from actions
Reading an order status is different from changing an address or approving a refund. Give each integration the minimum permissions it needs. Require the appropriate verification and approval for consequential actions, and log what happened.
When the bot does not know, it should say so and offer a useful next step. It should not invent a delivery date, policy exception or discount. Maintain an approved knowledge source and assign someone to update it.
Design the handoff as a workflow
Define escalation triggers: a direct request for a person, repeated misunderstanding, an unsupported request or a sensitive issue. Pass the relevant conversation summary and collected details to the team so the customer does not start again.
Tell the customer what happens next and use real support hours. Pause automated replies appropriately when a human takes over. Assign ownership of the queue; a handoff that nobody receives is only an error message with nicer wording.
Test the failure paths
Try a missing order, an unavailable integration, an ambiguous question and a customer who changes their mind. Confirm that retries do not create duplicate bookings or messages. Keep a manual fallback for outages.
Before production, review the current WhatsApp platform requirements and your messaging provider’s setup for the intended use. Platform policies and costs need their own validation. Measure resolution quality, escalation outcomes and customer complaints, not just the number of automated replies.