Overview
Your Digital Dermatologist: AI-Enabled Beauty Recommendation Engine
At - A- Glance
A comprehensive AI-powered skin analysis system that helps beauty shoppers on POINZON streamline and personalize decisions for more accurate and efficient product selection, skincare matching, beauty recommendations.
My Role
Product Design – User Research, Prototyping, and Interaction Design
Design System
Team
Product Manager x1
Product Designer x2
Content Designer x1
Front-end Eng x2
Back-end Eng x5
Timeline
Shanghai
Jun 2023 – Aug 2023
Tool
Figma, Miro, Adobe Suite
problem
It Is So Frustrating…
Today, beauty product selection processes that require personalization and precision, such as skincare matching and makeup shade selection, are still largely subjective, time-consuming, and prone to mismatch.
Data from customer reviews, expert recommendations, and marketing materials often conflict and overwhelm, while traditional online shopping platforms offer limited guidance for individual skin needs, leading to inefficient decisions and uncertain results. We see a clear opportunity to create a more intelligent, personalized, and confidence-building product experience to solve these beauty shopping challenges.
Today's Decision Making Progress
The current process is plagued by:
Poor self-diagnosis, limiting understanding of personal skin needs.
Overwhelming options across countless beauty brands and categories.
Color matching barriers when shopping makeup products online.
Conflicting information from reviews, experts, and brand marketing.
The solution
In Solving the Problem
Two components power our smart beauty matching system to improve shopping confidence and decision-making:
POIZON Beauty Product Match: AI-Powered Beauty Recommendations
Efficiency
Instant product matches, eliminating endless browsing time
Precision
Accurate skin-product matching to prevent incompatible purchases
Personalization
Recommendations tailored to each customer's unique skin profile
Expertise
Professional-level advice powered by beauty knowledge database
research
what are the challenges that customers facing when shopping beauty products at POIZON?
Survey responses
From in-app survey: “What challenges do you face when selecting beauty products?”
In-depth user interviews
Insights from our female beauty customers aged 18-35
EFFICIENT product discovery by matching specific needs, avoiding endless browsing and research.
Clear and TRANSPARENT ingredient information that directly addresses skin concerns.
PROFESSIONAL guidance ensures confident beauty product selection.
Competitive Analysis
Imagine having a personal beauty advisor that analyzes your unique skin profile to find your perfect product matches
AI skin test userflow
Product Recommendation userflow
Skin analysis feature
Pros
Comprehensive beauty product expertise with extensive feature sets.
Professional-grade skin analysis capabilities.
Cons
Complex user flows requiring redundant data entry.
No data persistence, forcing repeated skin information input.
Information overload creating inefficient shopping experience.
strategy
How might we make beauty product matching effortless and trustworthy?
Intuitive matching
Delivering precision-matched recommendations through AI skin analysis versus manual category browsing, eliminating trial-and-error searching.
Professional guidance
Provide expert application tips and ingredient compatibility insights alongside each suggested product.
Smart ingredient analysis
Decode complex ingredients through AI, filtering products to precisely target specific skin concerns while highlighting key actives and their proven effectiveness.
Personalized product discovery
Tracking progress and adapting recommendations as skin conditions evolve.
Design Exploration
Ideation-Rule Creation Flow
System Diagram
Wireframe
Design Development
Before
After
1. Multi-step questionnaire system that delivers more accurate skin analysis by understanding multiple aspects of each user's unique situation.
2. Three-step progressive question flow (gender → recommendation type → skin goals) reduces complexity while maximizing data quality.
Before
After
1. Changed infinite product list to no more than 6 recommendations, reducing decision fatigue.
2. Switched from vertical to horizontal scrolling, improving information density.
3. Added detailed recommendation page organizing products by Basics and Treatments , providing comprehensive guidance and building user trust.
4. Designed 'replace' feature to recommend alternative brands/formulations within the same product category, enhancing personalization flexibility.
Final Solution
Let's test your skin!
Scan facial data
In the beauty products detailed page, users can access the AI Skin Test feature. After enabling camera, they follow a face guide to take a selfie. To ensure personalized recommendations, users then complete a brief profile by indicating their gender, skincare goals, and product preferences.
Analyze: AI processes skin condition and generates report
This phase generates a skin report showing facial age and skin condition ranking. It presents skin insights through two tabs: Basic (skin tone and type) and Report Analysis (specific skin concerns). The baseline profile offers personalized care advice.
Recommend: Personalized product suggestions based on analysis
Products are organized by skincare routine steps with targeted symptom tags showing specific benefits. Users can customize their recommendations through a replace feature that offers alternative products while maintaining the same targeted benefits for their skin condition.
Design System
Design System
Consistent layout pattern flatten the learning curve
I implemented a cohesive set of visual and interaction patterns, unifying layouts by placing navigation in the top-left, content in the center, and actions to the right.
This design allows users to easily learn how to navigate, and stay focused on their tasks.
How to better organize complex contents?
I implemented a cohesive set of visual and interaction patterns, unifying layouts by placing navigation in the top-left, content in the center, and actions to the right.
This design allows users to easily learn how to navigate, and stay focused on their tasks.
Recommendation A/B Test
User feedback favored Concept B (eHorizontal Recommendations) as it's more concise to read, with a secondary tab for product replacement offering more efficient decision-making and time savings.
Test A
Test B ✓
Impact
User trust strengthens GMV potential
20%
Delivering precision-matched recommendations through AI skin analysis versus manual category
66.5% - 67.6%
Provide expert application tips and ingredient compatibility insights alongside each suggested product.
3M
Decode complex ingredients through AI, filtering products to precisely
Takeaway
Lessons I've learn…
The challenges of designing at scale
Designing at scale means multiples aspects with many constraints, such as data model complexity, evolving business priorities, user growth, and diverse use cases.
Each has different requirements and future adaptions. Moreover, it took time and effort to understand different stakeholders’ requirements and reach alignment. It was a humbling experience to learn that sometimes things take long for good reasons.