Cotton On - Retail Inventory Sales Forecasting

Project Overview

Cotton On Group is Australia’s largest global fashion retailer, operating across multiple countries with brands including Cotton On, Cotton On Kids, Cotton On Body, Factorie, Rubi, Typo, and Supré. We developed a pilot machine learning solution for 30 of their main Australian stores to predict inventory sales for 500+ products over rolling two week periods.

Key Contributions

Forecasting Model Development

• Developed and compared multiple time series forecasting approaches including Random Forest, SARIMAX, and XGBoost with lag features
• Built item-store level forecasting pipeline processing historical sales data across 30 stores
• Engineered temporal features including lag variables and trend indicators for model training
• Established evaluation framework measuring RMSE, MAE, and statistics for forecast quality

Data Analysis and Insights

• Analyzed sales patterns across 500+ products identifying consistent best sellers and seasonal trends
• Evaluated store performance hierarchies and cross-store demand patterns
• Generated two week rolling forecasts for item-store combinations
• Documented demand variability and identified high-confidence forecasting opportunities

Pilot Implementation

• Created forecasting utilities for item-store pair validation and filtering
• Implemented automated model training and diagnostic visualization pipelines
• Delivered exploratory analysis with recommendations for production deployment
• Established baseline forecasting capabilities demonstrating feasibility for scaling

Impact

The pilot project provided Cotton On with a foundation for inventory forecasting across their retail network. The exploratory analysis identified which products and stores had the most predictable demand patterns, with projections suggesting 70-90% accuracy potential for consistent items. The work established a technical baseline and roadmap for building production-ready forecasting systems, demonstrating clear value for future investment in predictive inventory management capabilities.