Dialog design is most of the bot's value. Step structure, intent training, and variable capture all influence how the bot feels to customers.
- Identify the intent the dialog handles
Each dialog corresponds to a specific customer goal. Write the goal in plain language; vague intents lead to dialogs that overlap and confuse.
- Sketch the conversation flow
Before opening Bot Builder, draft the conversation: bot greeting, what it asks, what it answers, what it does, what it confirms. The sketch becomes the dialog skeleton.
- Add intent training utterances
Provide 20 or more example phrasings per intent. Real customer language varies more than designers expect; thin training data produces brittle intent recognition.
- Build steps inside the dialog
Add Message, Question, Rule, Action, and Navigate steps in the order the conversation flows. Reuse common sub-dialogs through Navigate.
- Add Transfer to Agent fallback
Every dialog should end (or branch) with Transfer to Agent for cases the bot cannot resolve. The fallback is the difference between graceful handoff and customer frustration.
The customer goal that routes user input to this dialog.
The ordered Messages, Questions, Rules, Actions, and Navigate steps.
Context captured through Question steps and referenced later.
The work-doing steps (Flow, Apex, lookup, transfer) inside the dialog.
The fallback escalation when the bot cannot resolve.
- Monolithic dialogs become impossible to maintain. Break into small focused dialogs that compose through Navigate steps.
- Weak intent training (5 utterances per intent) produces brittle recognition. Provide 20 or more example phrasings per intent.
- Missing Transfer to Agent at the end of dialogs traps customers in dead-end conversations. Always design the escalation path.
- Repeated Question steps for the same variable produce a clunky bot. Capture once, reuse the variable everywhere downstream.