Countries vary sharply in readiness for AI-driven banking: nearly a third lead on either customer adoption or government AI capacity but not both; stronger government AI readiness correlates with deeper household digital-finance adoption after controlling for income, though causality is not established.

Main Finding

The paper constructs an AI-Banking Readiness Index (ABRI) that combines household digital-finance adoption (demand) with government AI-readiness (supply) for 97 economies and finds: (1) demand and supply readiness are only moderately aligned (r = 0.655), with 28 of 97 economies strong on only one side; (2) across countries, higher government AI readiness is positively associated with deeper household digital-finance adoption once income is controlled (β = 0.451, p = 0.017); (3) within countries over time, e‑government capacity (EGDI) is associated with account ownership in a way consistent with an infrastructure-mediated channel (suggestive, not a formal mediation test). Practically, the ABRI identifies each country’s binding constraint: supply-led systems need demand activation (connectivity, interoperable payments, digital literacy), while demand-led systems need AI governance and supervisory capacity before AI-driven personalized services can scale safely.

Key Points

• Contribution: Integrates demand-side financial-inclusion indicators with supply-side government AI-readiness into a single, diagnostic composite (ABRI). The emphasis is integrative and pragmatic rather than proposing a novel weighting approach.
• Index construction: Two-stage principal component analysis (PCA). First-stage PCs produce a demand sub-index (3 Findex indicators) and a supply sub-index (6 GARI pillars). Second-stage PCA on those two sub-indices yields the ABRI, rescaled 0–100.
• Core data:
  • Demand: World Bank Global Findex (2011, 2014, 2017, 2021, 2024 waves).
  • Supply (cross-section): Oxford Insights Government AI Readiness Index (GARI) 2025 (six pillars).
  • Supply (panel proxy): UN E-Government Development Index (EGDI) 2011–2024 used as time-varying proxy for state digital capacity.
• Methods beyond PCA: k‑means clustering, a demand–supply positioning matrix (to categorize economies as demand-led, supply-led, balanced, or low on both), cross-sectional regressions and two-way fixed-effects panel regressions, robustness checks.
• Samples:
  • ABRI computed for 97 economies (intersection of 2024 Findex and GARI 2025 coverage).
  • Cross-sectional regressions on 93 economies with full controls.
  • Panel: 639 observations across 142 countries (five Findex waves) after exclusions; a GARI-restricted AI-trend panel has 423 observations across 94 countries.
• Empirical controls: GDP per capita (income), connectivity measures (ITU), urbanization, institutional quality (WGI), financial depth, human capital.
• Diagnostics and caveats: GARI and EGDI are strongly correlated in the sample (r ≈ 0.82), motivating EGDI as a panel proxy, but EGDI is broader than AI readiness so panel results are interpreted as evidence on e‑government capacity. The paper flags composite-index fragility and treats channel evidence as suggestive rather than causal.
• Stylized policy result: 28 economies show single-sided leadership (good supply but weak demand, or vice versa), so supply-only rankings can conceal important mismatches.

Data & Methods

• Data sources:
  • World Bank Global Findex (2011–2024) — three demand indicators used: digitally enabled account ownership, digital payment activity, formal saving.
  • Oxford Insights Government AI Readiness Index (GARI) 2025 — six pillars as supply indicators.
  • UN E-Government Development Index (EGDI) 2011–2024 — used as time-varying supply proxy in panel work.
  • WDI, WGI, ITU — macro controls and connectivity variables.
• Index construction:
  • Standardize each indicator (z-scores).
  • For demand block: take first principal component of the three Findex indicators → demand sub-index; similarly, first PC of six GARI pillars → supply sub-index.
  • Rescale each sub-index to 0–100.
  • Take first PC of the two sub-indices and rescale to 0–100 → ABRI.
• Analytical methods:
  • Descriptive: correlation between demand and supply sub-indices, k‑means clustering and demand–supply matrix to classify country types, regional illustration (six Eurasian economies) to demonstrate diagnostic use.
  • Cross-sectional regression: demand sub-index (or digital finance outcomes) regressed on GARI and controls; key estimate β = 0.451 (p = 0.017) for government AI readiness effect conditional on income.
  • Panel regression: two-way fixed effects with EGDI and controls to examine within-country dynamics of account ownership over time; interprets a positive EGDI-account link as consistent with an infrastructure-mediated channel (not formally tested for mediation).
• Limitations noted by authors:
  • Unit of analysis is the country-level environment, not bank- or product-level AI deployment.
  • GARI is cross-sectional (2025); EGDI is a proxy with broader scope.
  • Channel evidence is suggestive; no formal mediation identification.
  • Composite-index choices (variables, PCA weights, rescaling) can influence results—authors prioritize transparency and robustness checks.

Implications for AI Economics

• Complementarity matters: AI-driven banking outcomes depend jointly on household digital-finance penetration and public-sector AI/infrastructure capacity. Research and policy should treat demand and supply as paired constraints, not as separate silos.
• Sequencing and prioritization:
  • Supply-led countries (strong GARI, weak demand): prioritize connectivity, interoperable payments, digital literacy and trust-building to create a market for personalized AI services.
  • Demand-led countries (strong demand, weak GARI): prioritize AI governance, supervisory capacity, data-protection, and public-sector AI competencies to ensure safe scaling of personalized, algorithmic services.
• Measurement and policy diagnostics: The ABRI provides a practical diagnostic for identifying binding constraints at the country level — useful for international development agencies, central banks, and regulators planning AI-enabled financial inclusion initiatives.
• Channels and causal questions: The finding that EGDI correlates with account ownership suggests an infrastructure-mediated channel (public digital capacity enabling digital finance), but causal identification remains open. Future work should:
  • Use quasi-experimental designs or instrumenting strategies to identify causal pathways (e.g., staggered rollouts of e-government platforms or broadband expansions).
  • Conduct formal mediation tests to separate direct supply effects (regulation, AI governance) from indirect infrastructure channels (connectivity, online services).
• Heterogeneity and distributional issues: The moderate alignment (r = 0.655) and 28 one-sided leaders underscore global heterogeneity; AI economics must account for varying complementarities across income levels, regions, and demographic groups (gender, age).
• Firm- and consumer-level implications: Macro readiness is necessary but not sufficient for bank-level AI adoption and consumer uptake. Micro-level frictions (trust, digital/financial literacy, preference for human fallback) remain crucial — suggesting combined supply-side reforms and demand-side interventions (training, disclosures, human-in-the-loop designs).
• Inequality and global impacts: The uneven distribution of ABRI across countries implies differential ability to capture the productivity and customer-experience gains from AI in banking, with implications for global financial inclusion and cross-country competitiveness in financial services.
• Research agenda: Extend ABRI-style integrative measures to (a) finer sectoral readiness (payment infrastructures, credit-data ecosystems), (b) longitudinal GARI measures or reconstructed AI-capacity indicators, and (c) link country-level readiness to bank-level outcomes (AI adoption, profitability, risk) and household welfare measures.

If you’d like, I can: - Extract the ABRI rankings for a specified set of countries (if publicly available in the paper’s tables), or - Draft short policy recommendations tailored to a specific country type (supply-led vs demand-led).