Digital innovation makes firms more resilient upstream but more dependent downstream: patents tied to digital technologies reduce supplier switching costs and spur supplier diversification, yet deep customer integration concentrates sales and creates an efficiency trap that weakens customer-side resilience.

Main Finding

Digital technology innovation has an asymmetric effect on firm supply-chain resilience. Using Chinese A‑share firms (2010–2022), the authors find that digital innovation strengthens supplier‑relationship resilience (promotes supplier diversification and makes switching easier) but weakens customer‑relationship resilience (promotes customer concentration via a “synergy efficiency trap”). Mechanism tests show the supplier benefit works through reduced supplier‑switching costs; the customer harm operates through deeper upstream–downstream integration that raises asset specificity and lock‑in.

Key Points

• Asymmetry: digital innovation acts like a centrifugal force vis‑à‑vis suppliers (encouraging multiple suppliers) but a centripetal force vis‑à‑vis customers (pulling business toward large, captive customers).
• Two channels identified:
  • Supplier‑screening/information channel: digital tools, platforms, data and distributed ledger features reduce search, monitoring and trust costs, facilitating supplier diversification and faster switching.
  • Product‑value/synergy channel: data‑driven customer insight, servitization, and deep ERP/API integration increase product‑customer fit and customer dependence; firms invest in customer‑specific digital assets, creating a positive feedback loop that raises switching costs and concentration.
• Mechanisms: evidence consistent with (a) lower supplier switching costs after digital innovation, and (b) a “synergy efficiency trap” where digital integration increases resource specificity and customer lock‑in, undermining customer‑side resilience.
• Measurement contribution: a new firm‑level digital technology innovation indicator constructed by matching corporate invention patents to a Classification Reference for Core Digital Economy Industries and the IPC (2023), intended to avoid shortcomings of word‑frequency measures in disclosures.

Data & Methods

• Sample: Chinese A‑share listed firms, 2010–2022.
• Key independent variable: firm digital technology innovation index built from matched patent data and CNIPA/industry classification (avoids “saying without doing” bias of text counts).
• Dependent variables: decomposed supply‑chain resilience into supplier‑relationship resilience and customer‑relationship resilience (operationalized via diversification/concentration metrics on upstream suppliers and downstream customers; paper treats resilience as the ability to maintain stability and recover after shocks).
• Empirical strategy: panel regressions with firm and year fixed effects (and industry fixed effects), robust standard errors. Baseline specifications regress supplier and customer diversification measures on the digital innovation measure and controls. Additional mediation/ mechanism tests examine supplier‑switching costs and measures consistent with increased synergy/specificity on the customer side.
• Robustness: multiple specifications and mechanism analyses to support causal interpretation (details in full paper).

Implications for AI Economics

• Dual externalities of digital/AI adoption: AI/digital tech generates opposite effects upstream versus downstream. Models of AI adoption and market structure should account for these asymmetric spillovers rather than treating supply‑chain partners symmetrically.
• Platform and lock‑in risk: AI‑driven data integration and servitization can create strong customer lock‑in and high sunk costs — relevant for market power, competition policy, and regulation of data portability/interoperability.
• Procurement and competition design: firms and policymakers should weigh supplier diversification benefits against customer concentration risks when promoting AI/digital investments; procurement platforms and standards that lower data‑specificity costs can reduce the “synergy efficiency trap.”
• Measurement note for empirical AI economics: patent‑based and classification‑matched indicators can be preferable to text‑frequency measures when assessing firm‑level digital/AI innovation, reducing measurement error from disclosures.
• Future modelling directions: incorporate asymmetric bargaining, sunk data/asset specificity, and endogenous switching costs into theoretical and empirical models of AI adoption, diffusion, and regulation.