

As the lead Interaction Designer for Xiaozhitiao — Kwai's first interactive feature — I spearheaded data-driven UX optimization, mastering how to leverage analytics insights to fuel iterative product enhancements.
Role
Interaction Designer
Duration
2024
Type

Kwai · Social Business
What is Xiaozhitiao?
Xiaozhitiao is Kwai's social interaction feature enabling users to exchange real-name and anonymous messages with mutual friends or strangers.
Whimsical message-delivery feature for friends
Adolescent social behaviors are driven by status display and emotional companionship


Anonymous chats and randomized message drops to strangers
Limited diversity in in-platform interactive mechanics


What did I do in the project?
Mapping user cognition
Define interaction paradigms
Through systematic audit of Kwai user-generated content, I identified dominant user cognition of Xiaozhitiao as an information carrier with higher-engagement.
User perception
Paper Plane
"Game"


User perception
XiaoZhiTiao
"Messaging"


Analyze feature-specific metrics
to drive core KPIs
Post-MVP launch, we leveraged funnel conversion rates and core action CTR analytics to refine page experiences.
UX Optimization
11
Message Send Rate
+7.87%
Home Feed Impressions
+13.86%
Negative Feedback
-65.75%
Harness cultural insights
to enrich feature experiences
User content analysis and interviews revealed collectible culture and style personalization as dominant trends among our Gen Z demographic.
This inspired the launch of customizable note themes.
MVP

Post-Optimization
Theme Previews




User Behavior Patterns



What did I learn in the project?

Let users choose a single dimension
Don't make the topic name semantic.
Pre-Optimization
Represent theme selections via textual buttons
Exposure of non-default theme is
2% - 10%
of default
Message volume is
10% - 50%
of the default

Post-Optimization
Prioritize visual theme previews over semantic labeling
Exposure of non-default theme is
37%
of default.
Average message volume is
57%
of the default


Maintain isolated single-variable manipulation
per experimental group
A single test group manipulated with two variables invalidates causal attribution
Isolating individual impact is statistically impossible

More…