P

Personalized Recommendation System

Built a recommendation system for Tesco at dunnhumby, identifying top personalized offers for millions of users. Leveraged collaborative filtering, user-item embeddings, and purchase history analytics on 100B+ retail transactions per year.

dunnhumby (Tesco Group)
2018-2022
Optimized personalized marketing and improved customer engagement

Project Overview

This large-scale recommendation system analyzed billions of retail transactions to deliver highly personalized product recommendations and promotional offers. The solution combined traditional collaborative filtering techniques with modern deep learning approaches to achieve superior recommendation accuracy.

System architecture: • Distributed processing of 100B+ annual transactions using PySpark • Hybrid recommendation model combining collaborative and content-based filtering • Real-time embedding generation for user and product representations • A/B testing framework for continuous optimization • Integration with Tesco's loyalty program and CRM systems • Automated offer generation and targeting

The system significantly improved customer engagement and purchase conversion rates through highly relevant personalized recommendations.

Key Challenges

  • Processing massive-scale transaction data efficiently
  • Maintaining recommendation quality during seasonal fluctuations
  • Ensuring real-time performance for online recommendations
  • Balancing personalization with business objectives

Outcomes & Impact

  • 25% improvement in recommendation click-through rates
  • 15% increase in personalized offer redemption
  • Enhanced customer loyalty and retention
  • Improved inventory turnover through targeted promotions

Technologies Used

PySparkCollaborative FilteringGCPEmbeddingsTensorFlow

Company

dunnhumby (Tesco Group)

2018-2022