The study, led by Shuang Zhang and Wei Wang of Harbin University of Commerce and affiliated institutions, set out to answer a question that has long hovered over agricultural digitalization: how does the mere presence of AI tools become something economically meaningful? Drawing on dynamic capability theory, the researchers treated smart agriculture not as a pile of sensors and algorithms but as an organizational digital-agriculture capability, the capacity of a firm or cooperative to integrate digital technologies into its operations and adapt them to shifting conditions. In their framework, AI adoption and transformational leadership are the inputs, smart agriculture is the mediating capability, and perceived regional economic benefits are the outcome respondents report.

The evidence base is a time-lagged survey conducted over roughly eight months, from May 2025 to December 2025, in Henan, Shandong, Jiangsu, and Sichuan provinces. In the first wave, approximately 820 questionnaires were distributed by email and in person, yielding about 589 usable responses. Respondents were then recontacted months later to report on smart agriculture and perceived regional economic benefits, a separation designed to reduce common method bias and clarify temporal ordering.

After rigorous screening that removed questionnaires with missing values, straight-lining, incompleteness, and duplicates, the final matched dataset contained 452 valid responses from owners, senior and farm managers, operations managers, digital transformation officers, and technical professionals across agricultural enterprises, agritech firms, cooperatives, and processing businesses. The analytical strategy combined two complementary techniques. Partial least squares structural equation modeling, or PLS-SEM, tested the network of associations among the latent constructs, using 5,000 bootstrap resamples to estimate significance.

Necessary condition analysis, or NCA, then asked a different question: whether minimum levels of any condition act as bottlenecks below which higher levels of the outcome become unlikely. The measurement model itself passed standard scrutiny, with item loadings between 0.812 and 0.941, Cronbach's alpha values from 0.920 to 0.939, average variance extracted between 0.715 and 0.869, and all heterotrait-monotrait ratios well below the conservative 0.85 threshold, confirming that the constructs, including the closely related pair of AI adoption and smart agriculture, are empirically distinct. The structural results were consistent across the board.

AI adoption showed the strongest direct association with perceived regional economic benefits, with a standardized coefficient of 0.307, and was also positively linked to smart agriculture at 0.176. Transformational leadership was positively associated with both smart agriculture, at 0.166, and perceived economic benefits, at 0.169. Smart agriculture itself was positively associated with perceived benefits at 0.140.

Crucially, the indirect paths through smart agriculture were statistically significant for both AI adoption and transformational leadership, supporting a partial mediation mechanism, though the indirect effects were small in magnitude. One of the most practically striking findings concerns human capital. Digital skills significantly strengthened the relationship between AI adoption and smart agriculture, with a significant interaction term of 0.125.

The moderation analysis showed that the association between AI adoption and smart-agriculture capability becomes stronger as respondents' digital competence rises, suggesting that sophisticated systems only pay off when people can interpret data, operate AI-enabled tools, troubleshoot problems, and translate algorithmic outputs into field-level action. In the effect-size analysis, digital skills contributed a moderate f-squared of 0.190 to smart agriculture, the largest practical contribution in the model, while most other direct effects were small. The model explained 34.3 percent of the variance in smart agriculture and 24.1 percent in perceived regional economic benefits, with adjusted values nearly identical, indicating stable explanatory power.

Predictive relevance checks supported the model's out-of-sample usefulness, with Q2-predicted values of 0.323 for smart agriculture and 0.224 for perceived benefits. An importance-performance map analysis further ranked the relative priorities of the constructs within the respondent-level model, and full collinearity tests, with variance inflation factors well below thresholds and a single-factor test explaining under 40 percent of variance, ruled out serious common method bias. The necessary condition analysis added a bottleneck perspective.

Transformational leadership emerged as a statistically significant necessary condition, with a permutation-based effect size of 0.081 and a p-value of 0.011, and AI adoption was also significant at 0.064 with p equal to 0.029. The bottleneck table showed that transformational leadership becomes a threshold at the 20th percentile of perceived benefits, AI adoption at the 60th, smart agriculture at the 80th, and digital skills at the 90th. In plain terms, respondents reporting the highest levels of perceived regional economic benefit were unlikely to do so when leadership or AI adoption fell below certain minimum levels, even if other conditions were favorable.

Digital skills and smart agriculture, by contrast, did not reach statistical significance as necessary conditions in the permutation analysis. The authors are careful, and repeatedly so, about what these numbers do and do not mean. Because the dependent variable captures respondents' assessments of broader economic benefits rather than measured regional GDP growth, agricultural value added, or employment, the results describe associations among perceptions at the individual level.

The time-lagged design separates measurement waves but does not by itself establish causality, and the study does not aggregate responses to the province level or model respondents as nested within organizations. The smart agriculture measure captures technology use and perceived adaptability rather than directly operationalizing the sensing, seizing, and reconfiguring processes at the heart of dynamic capability theory, a limitation the authors acknowledge explicitly. Even with those caveats, the study offers a coherent message for policymakers and agribusiness leaders: technology, leadership, organizational capability, and human skills move together in agricultural digital transformation.

Investments in AI infrastructure may be most effective when paired with leadership development and change-management training that encourage experimentation and coordination, and with digital literacy programs for farmers, managers, and technical staff. The authors call for future multilevel research that combines organizational data with objective regional economic indicators to test whether the perceptual relationships identified here correspond to realized economic outcomes, and for longitudinal designs that can track how AI-driven agriculture matures over time. For now, the picture emerging from China's agricultural heartland is one in which the smartest farm is not simply the most automated one, but the one where capable people, visionary leaders, and intelligent machines learn to work as a system.

Subject of Research: AI adoption, transformational leadership, and perceived regional economic benefits of smart agriculture in China

Article Title: Perceived regional economic benefits of smart agriculture: a dynamic capability and leadership perspective on AI adoption

Article References: Zhang, S., & Wang, W. (2026). Perceived regional economic benefits of smart agriculture: a dynamic capability and leadership perspective on AI adoption. Frontiers in Sustainable Food Systems, 10, Article 1865535. https://doi.org/10.3389/fsufs.2026.1865535

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1865535

Keywords: smart agriculture, AI adoption, transformational leadership, dynamic capability theory, digital skills, perceived regional economic benefits, PLS-SEM, necessary condition analysis, agricultural digitalization, China, agri-food organizations, mediation analysis