Using Artificial Intelligence to predict and optimize supplier lead times in procurement operations

Variability in supplier lead-times presents a considerable challenge for supply chain management as it creates late deliveries, lowered customer satisfaction, and higher costs of doing business. The objective of the research was to develop and test a machine learning model to predict supplier lead-time from a multi-source dataset, within the context of e-commerce. This dataset consisted of order, delivery, and supplier performance data. Using a quantitative, predictive model approach, a Random Forest Regressor model was trained to predict delivery lead time measured in days based on a variety of key operational factors, including, but not limited to, order volume, defective units, item category, and supplier reliability measures. The metrics used to measure model performance were Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and the coefficient of determination (R²). The Random Forest model achieved a mean absolute error (MAE) of 5.20 and root mean square error (RMSE) of 6.08 with a predictor metric R ² = −0.09, which indicates moderate predictive performance, and more optimal performance may be attainable with additional feature selection and potentially data collection. In terms of measure feature importance, Defective units, Average Lead Time from the supplier, and Supplier Lead Time Consistency could be evidence of the strongest predictors of delay. Overall, this study suggests that machine learning has the potential to provide insightful information relating to supplier performance patterns that can support procurement teams in auditing suppliers that are at moderate to heavy delay risk, which may improve forecasting ability and might direct the use of make-to-order inventory management capabilities to reduce delay and improve productivity in the supply chain.