Against the backdrop of intensifying global technological competition and the increasingly urgent need for self-reliance in core technologies across critical sectors, how enterprises can effectively leverage emerging technologies like artificial intelligence to drive breakthrough innovation has become a focal point for both academia and industry. Existing research predominantly examines AI's innovation effects from a technological determinism perspective, yet it fails to systematically uncover the complex internal and external organizational synergies essential for its efficacy. This study constructs an integrated analytical framework based on socio-technical systems theory. It examines how corporate AI applications (exploratory/exploitative AI applications) synergize with characteristics of the social subsystem,including the organization's digital transformation level, dynamic capabilities, executives' digital background, and the digital focus of the local government to collectively influence breakthrough innovation performance.
Using a sample of Chinese A-share listed manufacturing companies from 2020 to 2023, this study employs machine learning methods for empirical analysis. First, the K-Means clustering algorithm was applied to objectively classify enterprises based on the aforementioned socio-technical system characteristics, thereby identifying heterogeneous enterprise clusters that share similar intrinsic socio-technical configurations. Subsequently, the CART decision tree algorithm was employed to uncover the specific socio-technical system element combinations and decision rules that each cluster relies on to achieve high breakthrough innovation performance.
Key findings encompass two dimensions. First, according to the characteristics of socio-technical systems, sample enterprises can be categorized into three distinct groups: (1)Laggard enterprises exhibit low levels of both exploratory and exploitative AI applications, accompanied by slow progress in digital transformation, marked deficiencies in dynamic capabilities, inadequate internal and external organizational coordination, and thus overall stagnant development. (2)Pioneer enterprises demonstrate high levels of exploratory AI applications, digital transformation, and dynamic capabilities, along with strong forward-looking strategic awareness. However, their exploitative AI applications remain relatively weak, resulting in an unbalanced state characterized by “strong exploration, weak exploitation”. (3)Efficiency-driven enterprises demonstrate high levels of exploitative AI applications and actively respond to and leverage government digital policy support. Nevertheless, they tend to adopt a conservative stance in exploratory AI deployment, digital transformation deepening, and dynamic capability cultivation, thus following a development path centered on efficiency optimization and steady advancement.
Second, the attainment of high breakthrough innovation performance hinges on a socio-technical system configuration that matches the enterprise type. For laggard enterprises, if they possess a solid digital transformation foundation, active deployment of exploratory AI applications can effectively drive breakthrough innovation. Conversely, if their digital foundation is weak, the advancement of exploratory AI applications will instead inhibit breakthrough innovation. For pioneer enterprises, when their digital transformation foundation is robust, strengthening exploitative AI applications constitutes the key to overcoming breakthrough technological bottlenecks. If their digital capabilities are insufficient, they must rely on robust dynamic capabilities to advance exploratory AI applications. For efficiency-driven enterprises, when their digital foundation is weak but they receive strong government support, the synergistic advancement of both exploratory and exploitative AI applications can effectively foster innovation breakthroughs. In contrast, if they overemphasize exploitative applications to comply with short-term policy mandates despite having a mature digital foundation, they will risk hindering breakthrough innovation due to path dependency.
This paper makes four key theoretical contributions: refining the AI application classification framework by distinguishing the distinct roles of exploratory and exploitative AI in innovation;developing a novel enterprise classification paradigm based on socio-technical system synergies, offering a more nuanced theoretical lens for understanding context-dependent AI-enabled innovation; uncovering the multi-factor nonlinear configuration mechanism driving AI-enabled breakthrough innovation, thereby enriching research on multi-level interactive mechanisms of breakthrough innovation antecedents through a socio-technical systems perspective; and building an integrated analytical framework connecting "AI application-socio-technical system co-evolution-breakthrough innovation" , which extends socio-technical systems theory to the study of AI adoption and organizational innovation relationships.