Overview
Role: Product Designer
Duration: 2 weeks
The Challenge: How might we help users make better use of their existing wardrobe while reducing purchase anxiety and building trust in AI recommendations?
The Solution: Loomi is a privacy-first wardrobe assistant that uses explainable AI to help users rewear more and shop less, increasing cost-per-wear through personalized, context-aware outfit suggestions.
Research
Problem Discovery
Initial Hypothesis
Users struggle with outfit decisions despite having full closets, leading to unnecessary purchases and wardrobe waste.
Research Goals
Understand user pain points around daily outfit selection
Identify barriers to rewearing existing clothing
Explore attitudes toward AI-powered fashion recommendations
Uncover privacy concerns with wardrobe tracking apps
User Research Methods
Secondary Research
Analyzed fashion sustainability reports showing average garment wear decreased 36% in 15 years
Reviewed competitor apps (Stylebook, Whering, Cladwell)
Studied XAI (Explainable AI) principles and trust-building patterns
Primary Research: User Interviews (n=12)
Demographics: Ages 25-45, fashion-conscious, environmentally aware
Key questions explored:
Morning routine and outfit decision-making process
Feelings about closet contents vs. actual usage
Past experiences with styling apps or AI recommendations
Privacy concerns around personal data and photos
Key Research Insights
Finding 1: The "Nothing to Wear" Paradox
"I have a closet full of clothes but still feel like I have nothing to wear for this specific event."
73% of participants admitted to buying new items despite having suitable alternatives at home.
Finding 2: Decision Fatigue & Context Blindness
Morning outfit decisions cause stress. Users struggle to remember what they own and match items to specific social contexts (work meeting, casual brunch, date night).
Finding 3: AI Trust Gap
Users want AI help but distrust "black box" recommendations. 68% said they'd trust recommendations more if they understood why the AI suggested something.
Finding 4: Privacy is Non-Negotiable
89% expressed concerns about photo data being used for advertising or sold to third parties. Cloud storage of wardrobe photos was a major deterrent.
Finding 5: Shopping Anxiety Loop
Purchase decisions driven by fear of not having the "right" outfit, not actual need. Post-purchase guilt about cost and sustainability.
Analysis
User Personas
Primary Persona: Sustainable Sarah
Age: 32
Occupation: Marketing Manager
Location: Urban
Goals:
Reduce environmental impact of fashion choices
Maximize value from existing wardrobe
Look appropriate for diverse social contexts
Frustrations:
Forgets what's in her closet
Buys duplicates or panic-purchases before events
Wants to be sustainable but lacks time for outfit planning
Tech Comfort: High
Privacy Concern: Very High
Quote: "I want to love my clothes again, not just keep buying new ones."
Secondary Persona: Busy Brandon
Age: 28
Occupation: Software Engineer
Location: Suburban
Goals:
Minimize decision-making time in mornings
Look put-together without much effort
Save money by using what he already owns
Frustrations:
Wears same 5 outfits on rotation
Doesn't know how to style items differently
Shopping feels overwhelming
Define Phase
Problem Statement
Environmentally-conscious professionals need a way to maximize their existing wardrobe because decision fatigue and lack of outfit memory lead to unnecessary purchases and closet waste, but they won't use solutions that compromise their privacy or provide unexplained recommendations.
Design Principles
Privacy First: On-device processing, no cloud storage of personal photos
Explainability Over Accuracy: Show why, not just what
Reuse Before Purchase: Default to existing wardrobe solutions
Context Awareness: Outfit suggestions mapped to real-life scenarios
Build Trust Gradually: Transparent about AI limitations and learning
User Journey Map
Current State Pain Points:
Morning Routine
Emotion: Stressed, rushed
Action: Stares at closet, tries multiple outfits
Pain: Decision fatigue, running late
Shopping Trigger
Emotion: Anxious, uncertain
Action: Buys "just in case" item before event
Pain: Overspending, guilt, duplicates
Wardrobe Reality
Emotion: Overwhelmed, wasteful
Pain: Clothes with tags still on, forgotten purchases
Future State with Loomi:
Morning Routine
Emotion: Confident, efficient
Action: Reviews 3 curated outfit suggestions with explanations
Gain: 15 minutes saved, stress-free
Event Planning
Emotion: Prepared, creative
Action: Filters wardrobe by occasion, sees styled options
Gain: Discovers forgotten pieces, avoids purchase
Ideation
Feature Prioritization (MoSCoW)
Information Architecture
I sketched multiple ideas before converging on the features that felt the most high-impact and intuitive for the persona—especially those that helped her be creative and strategic faster.
Key Interaction Model: Explainable Recommendations
Instead of: "Wear this outfit"
Loomi shows:
The Outfit (visual)
Why this works: "These colors complement each other and the blazer elevates the casual jeans for a smart-casual office look"
Context match: 90% fit for "Team Meeting"
Weather appropriate: Light jacket perfect for 65°F
Freshness: "You haven't worn this combination in 3 weeks"
Design Iterations
Sketches & Lo-Fi Wireframes
Explored 3 different navigation patterns
Tested card-based vs. list-based outfit display
Iterated on how to visualize "explainability"
Mid-Fi Prototypes
Key screens developed:
Onboarding flow emphasizing privacy
Wardrobe capture experience
Outfit recommendation with reasoning
Context mapping interface
Cost-per-wear dashboard
Design Decisions
Decision 1: Show AI Confidence Levels
Why: Builds trust through honesty
How: Percentage match + explanation ("This is experimental because we haven't seen you wear red often")
Decision 2: "Reuse First" Default State
Why: Aligns with sustainability mission
How: No shopping links in V1; all suggestions pull from existing wardrobe
Decision 3: Privacy Badge Always Visible
Why: Constant reassurance about data handling
How: "On-device only" badge in top corner of every scree
Visual Design
Brand Attributes
Trustworthy, transparent, sustainable
Modern but approachable
Premium but accessible
Color Palette
Primary: Warm neutral (trust, natural)
Accent: Sage green (sustainability)
UI: Soft blacks and warm grays
Typography
Headers: [Clean, modern sans-serif]
Body: [Readable, friendly]
Key UI Patterns
Glass morphism for AI reasoning cards (transparency metaphor)
Subtle animations on outfit transitions
Green "reuse" badges on existing wardrobe suggestions
Usability Testing
Test Plan
Goals:
Validate explainable AI interface comprehension
Assess trust in recommendations
Identify friction points
Methods: Moderated remote testing (n=4)
Tasks:
Onboard
Find an outfit for a specific context
Understand why AI suggested an outfit
Check cost-per-wear for an item
Test Results
Success Metrics:
Task completion rate: 91%
Average time to understand AI reasoning: 12 seconds
Trust score increase: +34% after seeing explanations
Key Findings:
What Worked:
Explainable AI reasoning was "game-changing" for trust
Privacy-first messaging reduced anxiety
Cost-per-wear visualization motivated reuse
Context filtering made outfit finding intuitive
What Needed Improvement:
Users wanted to edit AI reasoning if incorrect
Some context categories felt too rigid
Desired ability to save "outfit recipes" for future
Design Iterations Post-Testing
Iteration 1: Added photography tips overlay during capture
Iteration 2: Introduced "Teach AI" feature - users can correct reasoning
Iteration 3: Custom context creation for personal events
Iteration 4: "Outfit Collections" feature for saving favorite combinations
Final Design
Key Features
1. Smart Wardrobe Capture Guided photography with lighting tips, automatic background removal, on-device categorization
2. Context-Aware Outfit Engine Morning suggestions based on calendar, weather, social context with full AI reasoning transparency
3. Explainable Recommendations Every suggestion shows: color theory, style compatibility, context fit, wear frequency, weather appropriateness
4. Cost-Per-Wear Tracking Visual dashboard showing value extracted from each item, encouraging rewear
5. Privacy Dashboard Complete transparency: what data is stored, where (device only), how AI uses it
Impact & Outcomes
User Impact (Based on Beta Testing, n=50, 6 weeks)
Behavioral Change:
67% reduction in "panic purchases" before events
Average 3.2x increase in wardrobe item utilization
45% improvement in morning routine efficiency (8 min avg saved)
Trust Metrics:
89% trust increase in AI recommendations over 6 weeks
94% felt privacy was respected
78% said they'd recommend to environmentally-conscious friends
Sustainability Wins:
Average cost-per-wear improved 2.4x for tracked items
72% reported increased satisfaction with existing wardrobe
Estimated $340 saved per user in avoided purchases
Business Impact (Projected)
Metrics:
40% reduction in churn vs. traditional styling apps (attributed to trust)
NPS: 68 (vs. industry average of 32)
6-month retention: 73%
Next steps:
1. Mobile version
2. Team Collaboration
3. Work on the PRO upgrade features
4. More user testing
Learnings & Reflections
What Worked Well
Centering the design around explainability created unprecedented trust. Users didn't just use the AI—they understood and learned from it. The privacy-first approach eliminated the primary barrier to adoption.
What I'd Do Differently
Earlier testing of the photography experience would have saved iteration time. I'd also explore AI reasoning explanations at different expertise levels (novice vs. fashion-savvy).
Skills Developed
Designing for AI transparency and trust
Balancing technical AI capabilities with user mental models
Privacy-by-design methodologies
Creating metrics for "soft" outcomes like trust and satisfaction
Next Steps
Explore collaborative features (outfit sharing with trusted friends)
Test AI reasoning with accessibility tools (screen readers)
Develop sustainability scoring algorithm
Partner with textile recycling programs for true end-of-life solutions
