Improving business process performance in SMEs through predictive modeling: a comparative study of statistical and machine learning models

Purpose This study investigates how predictive modeling can improve business process performance in small and medium-sized enterprises (SMEs) by enhancing demand forecasting. This paper examines statistical, machine learning and hybrid models to support process improvement by forecasting business outcomes, enabling data-driven decision-making. Using real-world data from a make-to-stock SME in the manufacturing sector, the research identifies context-aware forecasting strategies that align with business triggers and can be practically implemented without requiring extensive digital infrastructure. Design/methodology/approach A quantitative, comparative modeling approach is applied to real-world demand data from make-to-stock items, with a range of forecasting models evaluated using hyperparameter tuning. These models incorporate both endogenous demand trends and exogenous variables, and the results are critically assessed through a business process lens to evaluate practical relevance, scalability and workflow integration potential. Findings Hybrid and ensemble models, particularly Random Forest Regressor and Multi-Prophet, consistently outperform statistical approaches in forecasting non-linear, event-driven demand patterns. Feature-importance analysis confirms that episodic business events are stronger demand drivers than macroeconomic indicators, especially in project-based supply chains. Originality/value Drawing on operational data from an SME, this research moves beyond accuracy to focus on practical implementation, interpretability and process alignment. It positions predictive modeling as a decision-support subprocess embedded in SME operations, offering a replicable framework for data-driven forecasting in resource-constrained, real-world environments.