China’s supply-chain digitalization pilot nudged buyers toward AI-savvy suppliers: SCIAPP-designated firms increased the share of newly added suppliers with above-median AI capability by ~3.2 percentage points, an effect that strengthens over time and is largest for firms without substantial internal AI capacity.

Main Finding

Designation in China’s 2018 Supply Chain Innovation and Application Pilot Program (SCIAPP) causally increased firms’ tendency to add AI-capable suppliers: the share of newly added suppliers whose AI capability exceeds the industry-year median rises by about 3.2 percentage points. Event-study tests show no differential pre-trends and a steadily strengthening post-designation effect, consistent with gradual reorientation of procurement routines rather than an instantaneous or purely symbolic response.

Key Points

• Treatment and effect size: SCIAPP designation → +3.2 percentage points in the share of newly added suppliers above the industry-year median AI capability.
• Timing: No pre-treatment trends; effect grows over the post-treatment window, suggesting gradual changes in supplier selection practices.
• Moderation by firm AI orientation: The effect is weaker for firms with stronger pre-existing internal AI orientation, implying the program mainly nudges firms that had not already incorporated AI into supplier evaluation.
• Additional heterogeneity: Stronger program effects for firms with greater general digital capabilities, for non-manufacturing firms, and for smaller firms—contexts where external supplier AI capabilities matter more or are easier to leverage.
• Interpretation: Public supply-chain digitalization policy can shift interorganizational linkages by encouraging selection of technologically capable suppliers, extending the literature from internal capability upgrading to supplier-network composition.

Data & Methods

• Context: China’s 2018 SCIAPP, which designated firms as pilots in a staggered way across time/units.
• Sample: Panel of Chinese A-share listed firms, 2014–2023.
• Identification: Quasi-natural experiment exploiting staggered pilot design; difference-in-differences (DID) framework with event-study checks to assess parallel trends and dynamics.
• Outcome: Firm-level share of newly added suppliers whose AI capability is above the industry-year median (paper uses an AI-capability measure benchmarked at the industry-year level; see paper for exact construction).
• Robustness/auxiliary analyses: Event-study for pre-trends and timing dynamics; heterogeneity tests by firm internal AI orientation, general digital capability, sector (manufacturing vs non-manufacturing), and firm size.

Implications for AI Economics

• Policy efficacy: Targeted public digitalization programs can reallocate demand toward AI-capable suppliers, accelerating diffusion of AI capabilities across supply networks beyond firms’ internal adoption.
• Complementarity and substitution: Programs matter most for firms lacking strong internal AI—public policy can substitute for internal supplier-evaluation capacity and catalyze external sourcing of AI skills.
• Market structure and incentives: Shifts in buyer sourcing will alter demand signals to suppliers, potentially increasing returns to investing in AI capability and reshaping competition among suppliers.
• Heterogeneous impacts and targeting: Effects vary by firm size, sector, and digital maturity—policymakers should consider heterogeneity when designing interventions to maximize diffusion and equity.
• Dynamics and adjustment costs: The gradual rise in effect highlights path-dependent procurement routines; short-run ceremonial designation is insufficient—sustained support or complementary measures may be needed to embed new sourcing practices.
• Research directions: Study downstream productivity/welfare impacts, supplier investment responses, generalizability to non-listed firms and other institutional settings, and more granular measures of AI capability and supplier–buyer interactions.

Caveat: Findings are based on listed Chinese firms and an industry-year–benchmarked AI-capability outcome; generalization and the exact measure of AI capability should be checked against the paper’s measurement appendix.