Retail Shipping Prediction

The logistics team faced significant issues with the accuracy of their shipping estimates, which directly impacted workforce scheduling at distribution centers. The primary challenges were:

• Overestimating order picking leads to overstaffing, causing inefficiencies and increased operational costs due to unnecessary labor hours.
• Underestimating order picking leads to understaffing, causing shipping delays, stock distribution disruptions, and reduced customer satisfaction.

Enhance operational efficiency and cost-effectiveness by implementing a machine-learning model to accurately predict shipping volumes and automate shift scheduling.

• Develop and deploy a machine-learning (ML) model to accurately predict order picking volumes and automate shift schedules, ensuring optimal staffing levels.
• Replace existing manual and non-ML systems with the ML model to reduce operational costs and improve prediction accuracy.

We implemented a machine learning-based approach focusing on time series modeling. We evaluated ARIMA (AutoRegressive Integrated Moving Average), SARIMA (Seasonal ARIMA), and Prophet models, using metrics like MAPE (Mean Absolute Percentage Error), R-squared (R²), and MAE (Mean Absolute Error). Prophet emerged as the most effective model.

Sales Prediction

• Sales forecasting was initiated using the Prophet model
• The forecasted sales figures were integrated as regressors in the order-picking prediction framework

Model Selection

• Time series models (ARIMA, SARIMA, Prophet) were initially explored, aligning with the data's temporal dynamics
• Prophet was definitively chosen for its superior accuracy in forecasting and adept handling of seasonal and holiday variations

Evaluation Metrics

• Robust evaluation metrics such as MAPE, R² score, and MAE were employed to validate model performance
• Prophet consistently outperformed alternative models across these benchmarks

• Ability to identify and adjust for seasonal patterns and long-term trends in sales and order-picking data
• Predicted sales values are utilized as key inputs (regressors) in the order-picking prediction model, enhancing accuracy
• Combines sales predictions with other relevant features to forecast order-picking volumes
• Incorporates additional variables such as promotional events, holidays, and historical order volumes to improve prediction accuracy
• Regular updates and fine-tuning based on new data and feedback to maintain high accuracy
• Implemented automated processes designed to scale across multiple distribution centers, accommodating varying volumes and complexities
• Provides an intuitive dashboard for visualizing sales and order-picking forecasts, along with historical performance
• Capable of adjusting predictions in real-time based on new data inputs or changes in external conditions

• Improved accuracy in predictions: Enhanced forecast precision for sales and order picking, reducing forecasting errors.
• Optimized workforce scheduling: Ensured optimal staffing levels, minimizing labor inefficiencies and disruptions.
• Increased cost savings and efficiency: Minimized overstaffing and understaffing, reducing labor costs and operational disruptions.
• Enhanced decision-making and planning: Utilized data-driven insights for proactive decision-making and strategic planning.
• Improved customer satisfaction: Enhanced forecast precision for sales and order-picking ensured timely shipments, reducing delays and improving overall customer experience and loyalty.

• Python
• Pandas
• NumPy
• SciPy
• Prophet
• GCP - Vertex AI
• BigQuery
• GitHub