Product

Conversational Design Canvas

A seven-cell canvas for a chatbot or voice agent's persona, intents, and failure handling.

What it is

Seven cells: purpose and success metric, persona and tone, user intents ranked by frequency, sample dialogues for the top intents, context and memory, failure and fallback, and guardrails. Built so a bot's personality and edge cases get designed on purpose instead of by accident.

Reach for it when

• A support bot with an inconsistent tone and too many dead ends
• A new LLM-based assistant being scoped before any dialogue logic gets written
• A team that can describe what the bot should say but not what it should never say

Running it

1. Start with Purpose & Success Metric — tie it to a measurable outcome like containment rate or task completion, not 'be helpful.'
2. Define Persona & Tone next, since it constrains how every later dialogue should read.
3. Rank User Intents using real support or search-query data where you have it, then write 2-3-turn Sample Dialogues live as a group, reading them aloud.
4. Close with Failure & Fallback and Guardrails in the same 2-3 hour session — don't let time pressure cut these.