This study addresses the pressing challenges encountered by traditional foreign trade enterprises during the reconstruction of the global value chain. These challenges include issues such as data siloization, inefficient cross-border collaboration, and delayed responses to international market changes. To overcome these obstacles, a comprehensive digital foreign trade framework has been developed, which leverages advanced machine learning technology. This framework is designed to facilitate multi-modal data fusion and implement intelligent decision-making algorithms. The proposed solution outlines a strategic three-stage implementation path that consists of “data governance, decision optimization, and ecological collaboration.” Through rigorous empirical research, the findings reveal that the adoption of this innovative framework can enhance the efficiency of decision-making processes in foreign trade by an impressive 42.6%. Additionally, it significantly reduces the risk of misjudgment related to trade decisions by 31.8%. A key aspect of this framework is its innovative cross-domain feature fusion mechanism, coupled with a dynamic weight adjustment model. These components work together to deliver quantifiable and actionable decision support, which is critical for the successful internationalization of traditional enterprises. By integrating these advanced methodologies, the framework not only streamlines operations but also empowers businesses to respond more adeptly to the complexities of the global market.