Machine learning meets configurational analysis: unveiling multiple pathways to energy-saving innovation in construction firms

Purpose This study aims to systematically identify the key influencing factors and formation paths of energy-saving technology innovation, providing theoretical guidance and practical insights for construction enterprises to promote green innovation. Design/methodology/approach Based on the technology-organization-environment (TOE) framework, this study employs data from China's construction industry spanning 2017–2024. A mixed-method research design integrating machine learning (Random Forest), dynamic qualitative comparative analysis (QCA) and necessary condition analysis (NCA) is adopted to systematically explore the driving mechanisms of energy-saving technology innovation. Findings Random Forest analysis identifies investment in innovation resources, technical talent intensity, organizational slack, executives' IT background, government environmental subsidies and environmental regulation as core influencing factors. Necessity analysis reveals no single necessary condition, though investment in innovation resources, technical talent intensity and environmental regulation emerge as the earliest bottleneck constraints. Sufficiency analysis uncovers four equifinal paths: capability-driven and resource orchestration configurations (H1a and H1b) and institutionally-induced agile orchestration configurations (H2a and H2b), while also identifying governance misalignment configurations associated with inferior outcomes (L1a and L1b). Dynamic analysis shows that the overall consistency of all configurations remains stable over time, whereas their firm-level coverage varies substantially across firms. Research limitations/implications The study focuses on China's construction industry; future research could extend to other industries and national contexts. Practical implications Energy-saving technology innovation requires synergistic resonance of multidimensional factors. Enterprises should select differentiated innovation paths based on their resource endowments, while governments need to improve institutional environments to stimulate innovation vitality. Originality/value This study innovatively integrates machine learning with configurational analysis methods, revealing the causal complexity and asymmetry of energy-saving technology innovation and providing new perspectives for applying the TOE framework in the green innovation domain.