Demand forecasting and cost optimization for a chain of 8,000 convenience stores
A scalable and 100% reusable cloud solution that simplifies decision-making processes for 8,000 shops (supported with explainable models), reduces food waste and increases revenue.
Challenge
A leading chain of 8,000 convenience stores in Poland, with over 20 years of market presence, faced critical operational challenges affecting their supply chain management:
⚠️ Difficulty accurately forecasting sales of specific products for each store, time-frame, and day to ensure supply matched demand
⚠️ Need for a solution that business users could trust, with transparency into how predictions were made
⚠️ Requirement for a scalable, automated solution with optimized cloud costs and knowledge transfer to internal teams
Solution
🧪 End-to-End ML Pipeline: Built a comprehensive pipeline including data integration, validation, model tuning, evaluation, experiment tracking, and explainability
🔧 Custom Solution Development: Implemented custom loss functions to handle business requirements (penalizing underestimation more than overestimation)
🔄 Deployment Architecture: Created Docker containerized solution on Azure ML with automated scheduling and cost-optimized resource usage
🤝 Knowledge Transfer: Delivered dedicated workshops for the client's data science team, sharing industry best practices across machine learning and software development
Results
✅ Decision-making processes for 8,000 shops supported with explainable models giving confidence to business users
✅ Scalable and 100% reusable cloud ML pipeline adaptable to different product groups
✅ Reduced food waste and increased revenue through accurate demand forecasting
✅ Optimized computation time through parallelization with each product processing independently
✅ Cost-effective serverless infrastructure releasing resources after processing to minimize expenses
✅ Comprehensive knowledge transfer enabling the client's team to understand and maintain the solution
