Forecast Accuracy Improvement and Purchase Order Automation Using a Multi-Model Approach in Semiconductor Supply Chain: A Case Study

In semiconductor manufacturing, efficient procurement and inventory management are critical to minimize disruptions and ensure seamless production. This study presents a data-driven solution to address inefficiencies in purchase order generation and material forecasting through a custom-built 12-model ensemble forecasting algorithm. Unlike conventional methods, this framework blends statistical and machine learning techniques—ranging from time-series models to neural networks—to deliver highly accurate demand predictions. This paper focuses on per-item forecasting model selection development for semiconductor material planning. Unlike traditional approaches that apply a single model across all materials, semiconductor materials display highly heterogeneous consumption patterns due to differences in part types, usage rates, and production cycles. This heterogeneity makes a uniform forecasting approach suboptimal and motivates the need for per-item model selection. Rarely applied in the semiconductor domain, this ensemble approach enhances industry-specific forecasting and procurement practices. The model was trained and validated on real operational data from one of India’s largest semiconductor manufacturers, ensuring practical relevance and adaptability. Implementation results show marked improvements in forecast accuracy, reduced reliance on manual planning, and faster responses to demand-supply fluctuations, demonstrating a scalable path toward automation and resilience in high-tech procurement. The specific objective is to predict, for every relevant Part ID, the monthly material requirements needed to maintain uninterrupted production. This study was undertaken at a leading semiconductor manufacturing company in India, covering approximately 300–310 active material items sourced from 45–50 suppliers. The solution was integrated with the facility’s ERP system using Python-based automation and structured Excel imports for seamless execution.