AI-Driven demand sensing and lead-time risk modelling for U.S. mid-market manufacturers

The paper has touched upon the efficacy of classical forecasting and machine learning forecasting of demand sensing and lead-time risk modelling amongst the U.S. mid-market manufacturers. With the supply chains becoming more volatile and less transparent, operations become key to efficiently building demand. We offer a comparative analysis of a standard statistical modelling, i.e., Auto Regressive Integrated Moving Average (ARIMA) and Exponential Smoothing State Space Model (ETS), and the modern technique, i.e., Long Short-Term Memory networks (LSTMs). The study will explore both the merits and demerits of the two methods, especially when there is a paucity of data, and the demand trend is sporadic. Along these lines, the paper will discuss the use of external sources of data (such as port delays and supplier performance measurements) to increase the accuracy of the forecasting and decision-making process. Based on the benchmark data sets and simulated scenarios, we compare the performance of each model in terms of accuracy of prediction, efficiency in computational needs, and uncertainty resistance. Our findings have shown that, unlike machine learning models such as LSTM, which are powerful and effective, classical models are mostly easy to use and interpret, especially whenever dealing with complex and non-linear data. The paper culminates in a deployment playbook to offer practical insight to guide the mid-market manufacturers seeking to optimize their forecasting systems. The aspects of embedding AI-based models in ERP and supply chain systems also reveal how such models can be used to drive real-time decision-making.