For any new feature I'm specifying, I start with this simple process:
1. Map user decisions (not screens)
2. Identify context needed for each decision
3. Create the simplest possible representation
4. Get feedback on DECISIONS, not visuals
Let me break this down: