Designing chatbot scenarios that people complete

Choosing the first scenario

Look at what customers actually ask: support tickets, WhatsApp chats, call logs. Typical candidates:

• price and availability questions;
• booking or order placement;
• order status;
• collecting details for an enquiry before a manager calls.

Pick the one that is frequent, structured and valuable. A bot that handles one scenario reliably is better than one that tries ten and fails at most.

Mapping the dialogue

1. Entry: where the user comes from (website button, Instagram link, QR code) and what they expect.
2. Greeting and choice: a short welcome and the main options.
3. Steps: the questions needed to complete the task, in logical order.
4. Confirmation: a summary the user can check and correct.
5. Completion: what happens next — a booking confirmed, a manager will call, an order created.
6. Exits: handover to a person, restart, main menu.

Draw it as a flow with branches only where the answers really change the path.

Writing messages

• Short sentences; one idea and one question per message.
• Tell users what the bot can do and that a person is available.
• Use the tone of your brand, polite and clear.
• Avoid pretending the bot is human.
• Localise properly: Azerbaijani, Russian, English — each written natively, not machine-translated word for word.

Buttons or free text

• Buttons for choices with known options: services, dates, yes/no. They reduce errors and effort, especially on phones.
• Free text for names, addresses, descriptions of a problem.
• AI understanding for open questions or when users ignore buttons and write in their own words. See AI chatbot vs scripted bot.

A good bot accepts free text even when buttons are shown, and tries to understand it before asking again.

Validation and errors

• Check inputs such as phone numbers and dates, and explain problems politely.
• After two failed attempts, offer another way or a person.
• Handle timeouts: if the user leaves mid-scenario, remember progress or restart gracefully.
• Handle "stop", "menu", "operator" and similar commands at any step.

Handover

Define when the bot hands over: user request, repeated failure, sensitive topics, complaints, high-value opportunities. Pass the conversation history and collected data so the person does not ask again. See chatbot to human handover.

Integrations

Scenarios become useful when connected: availability from the booking system, order status from the order database, enquiries to the CRM. Design data needs together with the dialogue.

Test with real people

Before launch, ask colleagues and a few customers to complete the scenario without instructions. Watch where they hesitate or write something unexpected. After launch, review conversations weekly and improve. See measuring chatbot performance.