A collaborative project between Nova School of Business and Economics (Nova SBE) and Stokedge has developed an advanced predictive analytics solution aimed at optimizing inventory management for luxury retailers, focusing on forecasting weekly sales per brand and subcategory to address critical operational challenges in the retail sector.

A Luxury Collaboration

The team used advanced machine learning techniques to analyze retail data. Data scientists concentrated on feature engineering, model development, and validation, while software engineers focused on Cloud architecture and Big Data analytics. Beyond technical expertise in large scale data solutions, Stokedge provided operational and business insights, providing a comprehensive understanding of the complexities involved in luxury retail inventory management. This collaboration effectively bridged the gap between academic rigor and practical industry applications, addressing tangible challenges faced by luxury retailers.

Key Objectives

The project aimed to address three primary goals:

• Develop a multi-step forecasting model capable of predicting sales 5 weeks ahead
• Create a dashboard interface for retailers to monitor predictions and inventory levels
• Establish a pilot testing framework to validate model performance in production environments

These objectives were designed to provide luxury retailers with actionable insights for reducing overstock situations, improving seasonal inventory planning, and optimizing restocking operations.

Technical Methodology and Performance

The team implemented a rigorous development process, creating 22 engineered features, including price quartiles, seasonal indicators, and rolling sales metrics, and testing 8 regression approaches.

Support Vector Regression (SVR) emerged as the top-performing model, outperforming the baseline and alternatives like LightGBM and XGBoost. Temporal cross-validation was employed, along with learning curve and residual plot analyses, to ensure robust model performance. The solution architecture featured a data pipeline from Stokedge to AWS S3, through feature engineering and the SVR model, culminating in a predictions database feeding the retailer dashboard.

Outcomes and Business Impact

This collaboration project represents a significant technical achievement in applying machine learning to real-world business challenges. The Support Vector Regression (SVR) model demonstrated strong performance in predicting sales up to 5 weeks ahead across multiple product categories and brands, with particularly high accuracy for clothing and specific luxury labels. While the model’s accuracy decreases over longer prediction horizons, it still provides valuable directional guidance for inventory planning.

The expected business impact is substantial: local retailers can improve cash flow and reduce waste; large retail groups can enhance operational efficiency; and luxury brands can better align production with market demand. By bridging academic expertise with industry needs, this project sets a strong foundation for data-driven decision-making in the luxury retail sector, paving the way for smarter inventory management and more sustainable business practices.