Design System
- consumer
- AI
- maps
Discovery-first travel planning tool with AI-assisted itinerary building. Led product design from concept through public beta launch.
- Client
- Nomad
- Role
- Lead Product Designer
- Duration
- 6 months
- Year
- 2023
- Team
- 3 members — ML engineer, product manager, and senior frontend engineer
Overview
Nomad began as a response to how brittle existing travel planning tools are — they optimize for booking transactions, not for the exploratory phase where most decisions are actually made. We designed the experience around a map-first canvas where AI suggestions appear as lightweight overlays rather than interrupting the user's spatial reasoning. I led the end-to-end product design: journey mapping, IA, wireframes, visual design, and a three-month beta with 800 users. The beta validated a 34% improvement in itinerary completion rate compared to the leading competitor's linear flow.
Impact
- Itinerary completion
- +34%
- Beta cohort
- 800
- Day-7 retention
- 41%
vs. leading linear flow
active users over 3 months
of beta cohort
discovery
Discovery
Booking tools optimize for transactions, but most travel decisions happen during exploration. We needed a canvas where tentative plans could coexist with confirmed ones without forcing the user into a linear funnel.
Journey-mapping interviews with 14 frequent travelers surfaced a shared pattern: people open three to five tabs and toggle between them for days before booking anything. Linear funnels were collapsing that thinking time, not supporting it.
concepts
Concepts and testing
Suggestions appear as lightweight pins on the map, never modally. The model's confidence drives visual weight: high-confidence picks render solid, exploratory picks render as outlined ghosts that fade if ignored.
Two concept directions — a list-driven flow and the map-first canvas — were prototyped end-to-end and tested with 12 users on equivalent planning tasks. The map-first canvas won on both completion rate and self-reported confidence.
mockups
Mockups
testing
Final testing
An eight-week closed beta with 800 users validated the canvas pattern under real planning loads. Heatmaps showed users returning to the same map view across sessions — confirming the canvas as a persistent thinking surface, not a one-shot tool.
learnings
Challenges and Learnings
Calibrating AI confidence to visual weight took three iterations. Early versions over-trusted the model and rendered exploratory picks as solidly as confirmed ones, which users read as recommendations rather than ideas. Tying opacity directly to model confidence was the change that made the overlay feel honest.
“It's the first travel tool that doesn't punish me for changing my mind.”
next
Next steps
Public launch is targeted for late 2024, with a collaborative-canvas mode for couples and small groups in flight. The model retraining cycle is moving from quarterly to monthly to keep suggestion quality from drifting as user behavior evolves.