Artificial intelligence-driven governance threshold and sellers’ performance in cross-border e-commerce platforms

Background: Artificial intelligence (AI)-driven governance systems increasingly regulate seller decisions in cross-border e-commerce platforms. Mainstream research has treated AI-driven control intensity as a monotonically positive lever for platform coordination and seller outcomes. Aim: This study examines whether AI-driven governance intensity exhibits a non-linear performance boundary and identifies the mechanism under which governance expansion becomes counterproductive. Setting: The study was conducted among cross-border sellers on Amazon Global Selling, Alibaba International, and Shopee across China, Southeast Asia, and Eastern Europe. Method: Survey data from 207 cross-border sellers were analysed by using partial least squares structural equation modelling (PLS-SEM) with 5000-resample bootstrapping. Nonlinearity was confirmed by using a dual-condition criterion and a one-sided monotonicity test. Results: An inverted-U relationship emerged between AI governance intensity and seller performance (b_linear = 0.28, p = 0.021; b_quadratic = −0.19, p = 0.038), with a performance-reversing threshold at x* = 4.16 (bias-corrected and accelerated [BCa] 95% confidence interval: [3.9, 4.5]). Excessive governance suppresses independent strategic learning (b = −0.34, p < 0.001), partially mediating the performance penalty (indirect = −0.16; direct = 0.14, p = 0.046). Market volatility amplifies the post-threshold effect (b = 0.19, p = 0.014), with 23% of sampled sellers in the substitution regime. Conclusion: Governance intensity must be calibrated relative to a structural threshold beyond which automation displaces the strategic-learning capability on which competitive performance depends. Contribution: This study contributes a regime-contingent threshold model showing that AI-driven governance intensity and seller performance follow an inverted-U relationship, with a performance-reversing threshold at x* = 4.16. Strategic-learning suppression is empirically confirmed as the erosion mechanism (partial mediation, indirect effect = −0.16), and market volatility amplifies the post-threshold penalty, grounded in dynamic capabilities theory and offering platform designers a calibration boundary rather than a general prescription.