Notes for owners · AI and Intelligent Automation
How to Plan a WhatsApp Chatbot That Hands Over to Staff
Start by listing the questions staff answer every day, then decide which ones a bot must never take. Build short flows for the routine ones, with a route to a person at every step. GullySystem’s WhatsApp service plans handover first. A bot cannot answer from data your business has not recorded.
Ganesh HS, Strategy and Technology, GullySystem · · 3 min read
List the questions before drawing flows
Ask the people who answer the number to keep a tally for two weeks. Note each question, how often it came and where the answer was found.
Most lists repeat a few items. Order status, timings, list prices, payment details and appointment slots often lead. Those are the bot’s candidates.
- Question as the customer wrote it
- How often it came
- Where staff found the answer
- Whether it needed judgement
- Language the customer used
Decide what the bot must not handle
Agree this before writing a single flow. Complaints, refunds, credit, price negotiation and upset customers belong with a person.
Write the escalation triggers down. Words such as complaint or refund, two failed answers in a row, or a request for a person should all route to staff. The earlier messages travel with the chat.
Keep flows short and answers grounded
Long menus frustrate people. Aim for one or two turns per question, with quick-reply buttons where a choice is needed.
Answers come from the business’s own records. Order status is read from billing, slots from the booking system and prices from the approved list. Where the answer is missing, the bot says so and hands over.
AI replies need the same limit. An assistant drawing on approved documents should admit when a topic falls outside them.
Design the handover itself
A handover helps only if someone picks it up. Route conversations to a desk by topic or language, and show working hours.
Outside hours, tell the customer when a person will reply. Meta’s platform documentation says only approved templates can go out after a customer service window closes. Plan follow-ups around that.
- A desk or person for each topic
- Working hours shown to the customer
- Earlier messages visible to staff
- Internal notes hidden from the customer
- An approved template ready for late follow-ups
Pilot first, then read what the bot said
Run the bot with a small group of real customers and one team first. Read the conversations it handled. Look for wrong answers, loops and customers who gave up.
Review handover volume weekly. A topic that always reaches staff may need a better flow. Or it may simply belong with people.
Bot question register
A sheet listing each customer question, how often it comes, the data source for the answer and whether the bot or a person takes it. Add the escalation trigger for every row a person handles. The register becomes the specification for the flows.
Open a blank worksheet to printQuestions owners ask
Should a WhatsApp bot handle complaints?
No. Complaints need judgement and tone. The bot can collect the order number and route the chat to a named person.
Can the bot reply in Kannada or Hindi?
Yes. Templates can be approved in several languages, and replies can follow the language the customer writes in. Have a speaker check the wording before launch.
Do we need an AI model for a chatbot?
Not always. Fixed flows suit order status and timings. An AI assistant helps when questions vary, provided it answers only from approved documents.
Can the bot work if orders live only in chat threads?
Not on its own. A bot needs a billing, booking or order record to read from. Recording orders in a system comes first.
Sources
- Meta for Developers: About the WhatsApp Business Platform — verified 6 Oct 2026