It’s not AI! Designing for Automated Conversations
summary
Built two chat automations for ReferralCandy (5 years live) and new product Candy Bar, replacing some SMS flows and showing better engagement than SMS in the initial experiment. Design uses conversation trees and a written MVP script, not Sketch mockups; prototype created with Bot Society. Key lessons: label automation—not AI—don’t fake humanity, apologize and offer explicit fallback, and test constantly with quick guerrilla sessions (team: four designers, ~10 engineers, Amazon vouchers). Success metrics start vs SMS then shift to aligned business goals like referral revenue.
AI-written summary · about
chapters (18)
- 0:00Start
- 1:50Referral Assistant
- 3:41Candy Bar Assistance
- 7:00Alternatives to Sms
- 8:10Why Did We First Build Candy Bar Assistant
- 14:09Design Basics Are the Same but Tools Are Different
- 16:46Write Your Mvp Script
- 19:36Become a Really Good Writer
- 22:43It's Not Ai
- 24:09Automated Conversation Design
- 24:38Summary
- 25:03Design Basics
- 27:34How Does Your Team Measure Success with this Feature and Did You Meet Your Goals
- 29:44Constant Research and Constant Testing
- 30:41How Do You Avoid Users Frustration
- 32:00Build a Tone of Voice System like a Design System To Keep Your Messages Consistent
- 32:41Branding Workshop
- 32:57Branding
related talks
35mDesigning Experiences for Chatbots
50mThe Art of Coding a Conversation: Designing Bots
18mThe Art of Conversation (Design)
03mDesigning for humanlike interaction
46mDesign Patterns For AI Interfaces
39mHow Chatty Leads Builds AI Agents Businesses Actually Want
35mShipping with Confidence in the Age of AI
54mHow not to fuck up your bot UX