Design Conversational AI Features by Starting With the Conversation, Not the Spec
Design AI product experiences by mapping the actual conversation before writing requirements or building UI.
Outcome
A more realistic conversational AI feature spec grounded in user intent and failure paths.
Workflow steps
Write the ideal conversation
Script how the user and AI should interact when everything goes well.
Add failure paths
Map confusion, missing context, refusal, bad input, and escalation moments.
Turn conversation into requirements
Convert the conversation map into UX, data, policy, and evaluation requirements.
Why this workflow matters
Conversational AI features fail when teams specify screens before understanding the conversation. The interaction itself is the product surface.
How to run it
Start by scripting successful and failed conversations. Use those scripts to identify context needs, UI states, data dependencies, and guardrails.
What good looks like
The spec should describe user intent, AI behavior, escalation paths, and evaluation criteria.