The question is not whether to use a chatbot or people on WhatsApp but which conversations go where. Bots are fast, consistent and tireless; people understand, negotiate and apologise. This guide draws the line between them and shows how to hand over without frustrating anyone.
Read alongside this guide: WhatsApp Chatbot and Waki plans and pricing.
Key takeaways
- Bots for fixed answers and data collection; people for judgement, complaints and sales.
- Every flow needs a visible exit and an automatic handover after one failed attempt.
- AI replies from your own knowledge base extend the bot; they do not remove the human step.
- Review unanswered conversations weekly and move them to the right side.
What a bot does well
Anything with a fixed answer or a fixed next step.
- The first reply, at any hour, in seconds.
- Opening hours, location, price list, how to book, how to pay.
- Collecting the details a person will need: order number, preferred time, city.
- Status lookups when connected to your systems through a webhook.
- Opt-in confirmations, reminders and status updates triggered by events.
What a person does well
Anything where the answer depends on judgement or the customer is upset.
- Complaints, refunds and anything that starts with “I am not happy”.
- Sales conversations with questions the brochure does not answer.
- Negotiation, exceptions and promises.
- Anything medical, legal or financial that needs a qualified reply.
- Repairing a conversation the bot got wrong.
Designing the handover
Every flow needs a visible exit: a menu option that says “Talk to a person”, a set of trigger words such as complaint or refund, and a rule that routes anything unmatched after one attempt. The handover should assign the conversation to a named agent in the inbox with the context the bot collected, and the customer should be told it has happened and roughly when to expect a reply.
AI replies
AI answers from a knowledge base sit between a scripted bot and a person: they handle open questions in your own words, using your own OpenAI or Anthropic key. Give the AI a clear scope, tell it what it must not answer, review its conversations regularly, and keep the human handover one tap away. It is a very good first reader, not a replacement for your team.
A hybrid model that works
Let the bot take the first message and the predictable questions, and set a response-time promise for everything it hands over. Inside the 24-hour customer-service window your team replies freely, so a fast handover is also the cheapest path. Each week, read the conversations the bot could not finish and either add them to the flow or accept that they belong to people.
Signs the balance is wrong
Customers typing “human” or “agent” repeatedly, the same unanswered question appearing every day, agents re-asking what the bot already collected, or long gaps after a handover. Each one points to a specific fix: an exit option, a new branch, context passed on assignment, or more people at busy hours.
Next steps
Conclusion
Let the bot be fast and the people be human. Draw the line at judgement, make the handover obvious and quick, and revisit the line every week as the conversations teach you where it belongs.