Abstract: Research on data-driven decision-making methods for smart connected products (SCPs) has become increasingly diverse. However, most existing reviews focus on single application scenarios and lack a systematic framework that reveals the intrinsic logic and evolutionary patterns of cross-domain technologies. Based on the Web of Science and CNKI databases, this study selects more than 560 frontier publications from 2021 to 2025 as the core sample, complemented by a backward review of seminal early studies. Bibliometric and content analyses are conducted using the VOSviewer tool, with a focus on five representative categories of SCPs, including intelligent connected vehicles and robots, to systematically examine the technological evolution of data-driven decision-making methods. The results indicate that: 1) SCP decision-making methods follow a three-stage evolutionary trajectory jointly driven by data and scenario complexity, progressing from early machine learning approaches based on static perceptual data, to reinforcement learning methods based on dynamic interactions, and further to advanced intelligent decision-making approaches for complex scenarios; 2) bibliometric evidence shows that intelligent connected vehicles, unmanned aerial vehicles, and robots are the most concentrated application domains of data-driven methods, with reinforcement learning serving as a key enabling technology that links different application scenarios and significantly enhances the autonomy of SCPs; 3) current research faces three major bottlenecks, namely data dependency, trustworthiness, and collaborative ecosystems, and future studies should focus on data-knowledge integration, lifecycle-wide trust assurance, and the evolution of human-machine collaborative intelligence. This study reveals the underlying mechanisms of data-driven decision-making for SCPs and provides theoretical foundations and methodological references for technology selection and industrial intelligent upgrading.