Machine Learning

🎯 SITUATION

The Digital Shopping Divide (2020)

Imagine walking into a Currys store where knowledgeable sales professionals greet you, understand your needs, and suggest the perfect laptop case for your new MacBook or the right HDMI cable for your TV setup. Now imagine going online and... nothing. Just static product pages with no guidance, no suggestions, no help.

This was the reality facing Dixons Carphone (now Currys PLC) as customer shopping behaviors evolved rapidly. The pandemic had accelerated the shift to online shopping, but our digital experience felt lifeless compared to the human touch customers received in-store.

The Problem Landscape:

• Fragmented Shopping Journeys: Customers shop online, in-store, reserve and collect—but each experience felt disconnected
• Lost Revenue Opportunities: No intelligent cross-selling meant missing out on natural product pairings
• Static Digital Experience: Websites showed products in isolation, unlike the helpful bundling guidance available in stores
• Competitive Disadvantage: Other retailers were moving toward personalized experiences while we remained static

‍

The Data Wake-Up Call

Our analytics revealed the stark reality:

• Low cross-sell rates on digital platforms
• High cart abandonment when customers couldn't find complementary products
• Missed opportunities for incremental sales and margin growth
• Customer frustration with impersonal online experiences

The Vision: Transform our static digital catalog into an intelligent recommendation engine that could replicate—and enhance—the personalized service customers received from our skilled retail staff.

📋 TASK

Building the future of personalized retail

The Mission Brief

I was tasked with collaborating on the design and user experience strategy for "Supercharge Attach"—an ambitious project to build a machine learning-powered recommendation engine that would:

Primary Objectives:

• Replicate In-Store Expertise: Bring the knowledgeable sales assistant experience to digital channels
• Drive Incremental Sales: Increase cross-sold products and basket attachment rates
• Personalize at Scale: Deliver real-time, individualized product suggestions to millions of customers
• Unify the Experience: Create consistency between online and in-store shopping journeys

The Technical Challenge

This wasn't just a UX project—it required designing the interface between complex machine learning systems and human behavior:

System Requirements:

• Real-time API integration with product detail pages
• Seamless fallback from personalized to natural recommendations
• Scalable architecture within Google Cloud Platform (GCP)
• Integration with Syntasa's behavioral data synthesis

User Experience Goals:

• Make recommendations feel helpful, not pushy
• Maintain browsing flow while introducing relevant suggestions
• Balance personalization with privacy concerns
• Design for both new visitors and returning customers

The Dual Challenge

For New Visitors: Create compelling "natural attach" recommendations based on product relationships

For Returning Customers: Leverage browsing history to deliver highly personalized suggestions

Success Metrics Defined:

• Product coverage increase across the website
• Add-to-basket rate improvements
• Cross-sell conversion rates
• User engagement with recommendation features

⚡ ACTION

Engineering personalization that actually works

Phase 1: Understanding the Ecosystem (Weeks 1-4)

Mapping the Customer Journey

Working with the data science team, I mapped how customers actually shop across our digital ecosystem:

• Product discovery patterns
• Cross-category browsing behaviors
• Abandonment points in the purchase funnel
• Moments where recommendations would add value vs. create friction

Competitive Analysis Deep Dive

I studied how industry leaders like Amazon, Best Buy, and Apple integrate recommendations:

• Placement strategies that felt natural
• Visual treatment that enhanced rather than cluttered
• Personalization approaches that built trust
• Fallback systems for when data was limited

Phase 2: The Recommendation Engine Architecture (Weeks 5-8)

Designing the Intelligence Layer

Collaborated with Syntasa to understand the technical foundation:

For Personalized Recommendations:

• Nearest Neighbor model identifying similar customers
• Behavioral data synthesis within GCP environment
• Real-time API queries based on visitor ID and product ID

For Natural Attach Recommendations:

• Non-personalized algorithm for new visitors
• Product relationship mapping based on purchase patterns
• Fallback system ensuring every visitor sees relevant suggestions

User Experience Flow Design

I designed the recommendation experience to feel seamless:

User Journey Flow:
├── Product Page Visit
├── Real-time API Query (visitor ID + product ID)
├── Intelligent Routing:
│   ├── Sufficient History → Personalized Recommendations
│   └── New/Limited History → Natural Attach Recommendations
└── Dynamic Display of Relevant Bundles


Phase 3: Interface Design & Testing (Weeks 9-16)

Creating the Recommendation Components

Designed multiple recommendation interface patterns:

Bundle Suggestions:

• Visual product groupings that felt natural
• Clear value proposition for each bundle
• Easy add-to-cart functionality for individual items or bundles

Cross-Sell Placements:

• Strategic positioning that enhanced rather than disrupted browsing
• Progressive disclosure to avoid overwhelming users
• Mobile-first responsive design

A/B Testing Strategy

Implemented comprehensive testing methodology:

Test Scenarios:

• Recommendation placement positions
• Visual treatment variations
• Personalized vs. natural recommendations performance
• Bundle presentation formats

Key Metrics Tracked:

• Click-through rates on recommendations
• Add-to-basket conversion improvements
• Overall page engagement metrics
• Revenue per visit increases

Phase 4: Productionization & Optimization (Weeks 17-24)

Technical Implementation

Working with Syntasa's Composer module, we productionized the solution:

• Seamless integration with existing Currys infrastructure
• Real-time performance monitoring
• Automated failover systems
• Scalable architecture supporting millions of daily queries

Continuous Learning Integration

Designed feedback loops to improve recommendations over time:

• Purchase behavior analysis
• Recommendation effectiveness tracking
• Dynamic model updates based on seasonal trends
• Cross-channel data integration

🏆 RESULT

Transforming digital commerce through intelligent personalization

Quantitative Success Metrics

📈 Revolutionary Product Coverage

Before: 32% coverage with manual bundling

After: 72% coverage with AI-driven recommendations

Impact: 2x increase in product coverage across the website

This meant customers now encountered helpful product suggestions on nearly three-quarters of their browsing journey, compared to less than one-third previously.

🛒 Dramatic Conversion Improvements

Personalized Recommendations: 3x higher add-to-basket rate

Natural Attach (for new users): 1.3x improvement over manual recommendations

These weren't marginal gains—they represented fundamental improvements in how customers engaged with our product ecosystem.

Qualitative Transformation

Customer Experience Revolution

Before: "I didn't know I needed a laptop case until I got home"

After: "The website suggested exactly what I needed—just like talking to a helpful store employee"

Business Intelligence Gains

• Real-time Insights: Understanding customer preferences as they browse
• Inventory Optimization: Data-driven understanding of product relationships
• Marketing Intelligence: Personalization data informing broader campaign strategies

The Ripple Effect Across the Business

Engineering Team Empowerment

The project created a scalable foundation for future personalization initiatives:

• Reusable machine learning infrastructure
• Proven methodology for A/B testing recommendation algorithms
• Cross-functional collaboration model for data science projects

Customer Relationship Evolution

• Trust Building: Helpful suggestions that genuinely added value
• Engagement Increase: Customers spending more time exploring relevant products
• Purchase Confidence: Reduced anxiety about missing complementary items

Long-term Strategic Impact

Data-Driven Culture Shift

This project proved the value of machine learning in retail, leading to:

• Increased investment in personalization technology
• More sophisticated customer behavior analysis
• Integration of ML insights into broader business strategy

Scalable Personalization Platform

Built infrastructure supporting future innovations:

• Cross-channel recommendation consistency
• Advanced segmentation capabilities
• Real-time personalization across the entire customer journey

💡 KEY LEARNINGS & IMPACT

What this transformation taught about AI-driven design

Major Design Insights

1. Personalization Must Feel Natural

The most successful recommendations felt like natural discoveries, not algorithmic pushes. Design played a crucial role in making AI feel human.

2. Fallback Systems Are Critical

New users still received valuable recommendations through our "natural attach" algorithm. Never leave any user without helpful suggestions.

3. Real-time Matters

The ability to serve recommendations in real-time based on current browsing behavior created significantly higher engagement than static suggestions.

4. Context Is King

Recommendations that understood the customer's current journey stage (browsing vs. ready to buy) performed dramatically better.

The Bigger Business Story

This project demonstrated how thoughtful UX design can amplify the power of machine learning to create genuine business value. By focusing on the human experience of AI-powered recommendations, we achieved:

• Customer Satisfaction: Genuinely helpful suggestions that enhanced shopping
• Business Growth: Measurable increases in cross-sell and basket values
• Technical Excellence: Scalable platform supporting future innovations
• Cultural Shift: Proof that data science and design together create magic

Future Implications

The success of this project opened doors for:

• Advanced personalization across all digital touchpoints
• Cross-channel recommendation consistency
• AI-powered customer service enhancements
• Predictive inventory management based on customer behavior

The Ultimate Learning: The most powerful AI is invisible to the user—it simply makes their experience better, more helpful, and more human.

🛠️ Skills Demonstrated

• AI/UX Integration: Bridging complex machine learning systems with intuitive user experiences
• Data-Driven Design: Using behavioral analytics to inform design decisions
• Cross-functional Collaboration: Working seamlessly with data scientists, engineers, and product teams
• A/B Testing Methodology: Systematic validation of design hypotheses
• Systems Thinking: Understanding how personalization fits into broader customer journeys
• Scalable Design: Creating solutions that work for millions of users
• Business Strategy: Connecting user experience improvements to measurable business outcomes

This case study showcases how strategic UX design can transform machine learning technology into meaningful customer value and business growth.