Retailers facing ambiguous future markets are more likely to adopt advanced digital tools, but familiar financial, technological and institutional risks deter uptake and blunt that incentive.

Main Finding

Uncertainty about future market and ecosystem conditions can stimulate advanced digital technology adoption in retail firms, but identifiable risks (financial, technological, institutional) reduce adoption and weaken the positive effect of uncertainty. Adoption occurs through multiple firm- and context-specific pathways, and firms need sufficient capacity (size, innovation intensity, finance access, ecosystem embeddedness) for uncertainty to translate into adoption.

Key Points

• Sample: 1,256 retail firms across EU Member States.
• Conceptual distinction: risk = identifiable adoption barriers; uncertainty = ambiguity about future market/ecosystem conditions.
• Core empirical results:
  • Risk is negatively associated with advanced digital technology adoption.
  • Uncertainty is positively associated with adoption.
  • The interaction between risk and uncertainty is negative: identifiable risk attenuates the stimulative effect of uncertainty.
• Predictors confirmed as important: uncertainty, risk, firm size, innovation intensity, and contextual conditions.
• Heterogeneity and multiple paths: fsQCA shows several configurational pathways to adoption involving combinations of risk, uncertainty, firm size, finance access, digital growth orientation, and ecosystem embeddedness.
• Methods triangulation: classical regressions (logit, probit), machine-learning (XGBoost, LightGBM) with SHAP for interpretability, and fsQCA for configurational causality.
• Machine-learning models achieved predictive performance similar to regression models and helped confirm variable importance via SHAP values.

Data & Methods

• Data: cross-sectional firm-level survey data from 1,256 retail firms in EU Member States.
• Dependent variable: adoption of advanced digital technologies (binary).
• Key independent variables: measures of financial, technological and institutional risk; measures of market/ecosystem uncertainty.
• Statistical approaches:
  • Econometric: logit and probit models including interaction terms between risk and uncertainty.
  • Machine learning: XGBoost and LightGBM for prediction; SHapley Additive exPlanations (SHAP) to interpret feature contributions.
  • Configurational analysis: fuzzy-set qualitative comparative analysis (fsQCA) to identify multiple sufficient configurations leading to adoption.
• Robustness: consistency between regression and ML predictive performance; complementary fsQCA highlights equifinality and causal complexity.

Implications for AI Economics

• Reframing uncertainty: Uncertainty need not always be a barrier—when firms have capacity, uncertainty can incentivize investment in advanced technologies (including AI). Models of technology adoption should allow for positive effects of uncertainty, not only negative risk effects.
• Heterogeneity matters: Policy and firm-level interventions should account for multiple adoption pathways. One-size-fits-all incentives may be ineffective; tailor support by firm size, innovation capability, finance access, and ecosystem embeddedness.
• De-risking is crucial: Reducing identifiable barriers (finance, standards, technological support, institutional clarity) amplifies the positive influence of uncertainty on adoption—policies that lower these risks (subsidized finance, clear regulations, technical assistance) can unlock adoption motivated by uncertain future opportunities.
• Measurement and modeling recommendations:
  • Include both risk and uncertainty measures and their interaction in empirical models of AI adoption.
  • Use interpretable ML (e.g., SHAP) alongside econometrics to validate predictors and improve prediction.
  • Employ configurational methods (fsQCA) to capture multiple sufficient pathways and complementarities between firm capabilities and contextual conditions.
• Targeting investments: Support smaller or resource-constrained firms by strengthening finance access, innovation capacity, and ecosystem ties so they can convert market uncertainty into adoption rather than being stifled by risk.
• Future research directions: longitudinal studies to identify causal dynamics over time, sectoral comparisons for AI-specific adoption patterns, and evaluation of targeted de-risking policies on AI uptake and productivity.