Bank staff report that stronger regulation, better tech readiness and internal capacity are linked to more efficient credit analysis; technological readiness shows the largest association, explaining about 67% of variation in perceived efficiency. The evidence is correlational and based on a small, single-country survey, so claims about causal impact are limited.

Main Finding

Machine learning (ML) adoption determinants—regulatory pressure, technological readiness, and organizational capacity—are all positively and significantly associated with perceived efficiency of credit analysis in Kenya’s Tier II commercial banks. Technological readiness had the largest measured association (r = 0.819; R² = 0.670), followed by organizational capacity (r = 0.803; R² = 0.645) and regulatory pressure (r = 0.616; R² = 0.379). The authors conclude that strengthening technology, organizational capabilities, and supportive regulation can materially improve credit-analysis efficiency in resource-constrained banks.

Key Points

• Study and authors: Catherine Muthuuri & Allan Kihara (Chandaria School of Business, USIU-Africa), International Journal of Finance, Vol.10 Iss.8 (2025).
• Context: Focus on Tier II commercial banks in Kenya (mid-sized banks serving SMEs and low-income borrowers) where ML adoption lags Tier I banks.
• Theoretical framing: Technology–Organization–Environment (TOE) and Diffusion of Innovation (DOI) theories to explain adoption drivers.
• Sample: Census of 48 professionals across eight Tier II banks (credit risk managers, data analysts, loan officers, IT/digital officers, compliance officers, strategy leads).
• Measurement: Primary data via structured questionnaires (5‑point Likert); outcomes are perceptions of efficiency of credit analysis; predictors cover regulatory pressure, technological readiness, and organizational capacity.
• Main statistical results:
  • Regulatory pressure: Pearson r = 0.616 (p < .05); regression R² = 0.379; F(1,46)=28.059; β = 0.616 (p < .05).
  • Technological readiness: Pearson r = 0.819 (p < .05); regression R² = 0.670; F(1,46)=93.375; β = 0.819 (p < .05).
  • Organizational capacity: Pearson r = 0.803 (p < .05); regression R² = 0.645; F(1,46)=83.527; β = 0.803 (p < .05).
• Descriptive insights: Respondents generally agreed that CBK guidelines, ICT compliance, data-protection rules, and infrastructure (internet/power backup) influence ML adoption and credit-analysis modernization; some variability in perceptions across banks/roles.
• Recommendations (authors): Central Bank of Kenya should refine regulations that encourage ML; Tier II banks should invest in digital infrastructure (including cloud), data quality, cybersecurity, and human capital (recruitment, training, leadership/culture change).

Data & Methods

• Design: Descriptive, cross-sectional survey.
• Population & sampling: Targeted professionals involved in credit risk and digital initiatives at Tier II banks; census sampling used given small population (N = 48).
• Instrument: Structured questionnaire with Likert-scale items measuring:
  • Regulatory pressure (CBK guidelines, risk-based lending requirements, ICT/compliance, data protection, supervisory pressure).
  • Technological readiness (infrastructure, cloud/mobile integration, cybersecurity) — reported as a composite predictor in analyses.
  • Organizational capacity (human capital, leadership support, training, change management) — composite predictor.
  • Outcome: Efficiency of credit analysis (perceived improvements from ML adoption).
• Analysis: Descriptive statistics, Pearson correlation, and simple/multiple linear regressions run in SPSS v28.0; significance threshold p < .05. Regression model specified as Y = β0 + β1X1 + β2X2 + β3X3 + ε (Y = efficiency; X1 = ML adoption level; X2 = technological readiness; X3 = organizational capacity).
• Limitations noted (implicit in methods and results):
  • Small sample, single bank tier, and cross-sectional self-reported data limit external validity and causal inference.
  • Potential common-method bias (same respondents provided predictor and outcome measures).
  • Outcome is perceived efficiency rather than independently observed performance (e.g., measured NPL reductions or decision times).

Implications for AI Economics

• Productivity and credit-market outcomes
  • Large explanatory power of technological readiness (67% of variance) and organizational capacity (64.5%) suggests investments in infrastructure and skills can yield substantial efficiency gains in credit assessment—potentially lowering loan processing costs and improving pricing/credit allocation.
  • Improved credit-analysis efficiency may reduce non-performing loan rates for Tier II banks, influencing bank profitability, credit supply to SMEs, and financial inclusion dynamics.
• Adoption drivers & policy design
  • Regulatory pressure explains a notable share of variance (37.9%). Well‑designed regulation and RegTech incentives can accelerate ML uptake—but regulation should be coupled with capacity building (infrastructure and skills) to avoid compliance burdens that do not translate into efficiency gains.
  • Policymakers (e.g., central banks) can use a mix of standards, sandbox environments, and targeted subsidies/partnerships to mitigate fixed-cost barriers for mid-tier banks.
• Market structure and competition
  • If mid-tier banks successfully adopt ML, competitive dynamics could shift: Tier II banks may regain market share vs. digital lenders and large banks, affecting interest spreads and lender heterogeneity.
  • Heterogeneous adoption can lead to winner–loser dynamics across banks; modeling adoption as a function of fixed capital/skill constraints is important for macroprudential and competition analyses.
• Labor and human-capital effects
  • Findings underscore the importance of organizational capacity—implicating upskilling and reallocation of labor within banks. Economics models should incorporate training costs, labor reconfiguration, and potential displacement vs. productivity complementarities from ML.
• Research and measurement suggestions for economists
  • Move beyond perceptions: use transactional data to measure effects of ML on default prediction accuracy, approval rates, processing times, and NPL trajectories.
  • Causal identification: exploit staggered rollouts, regulatory changes, or randomized pilots to estimate causal effects on loan outcomes and bank performance.
  • Heterogeneity and general equilibrium: estimate how differential adoption across bank sizes affects credit supply, interest rates, and SME growth at the regional/national level.
  • Cost–benefit/ROI: quantify upfront investment, operating costs (cloud, data management, cybersecurity), and benefits (reduced provisioning, higher throughput, reduced loss rates).
• Potential risks to incorporate into economic models
  • Data quality and governance constraints can limit realized gains; models should account for noisy data environments that reduce ML effectiveness.
  • Cybersecurity and privacy compliance impose costs that may alter net returns to ML investments.
  • Regulatory uncertainty can delay adoption; economic policy simulations should include regulatory clarity and enforcement regime scenarios.

Suggested next empirical steps for the literature - Replicate with larger, multi-tier samples and objective performance metrics (NPLs, accuracy, time-to-decision). - Longitudinal or quasi-experimental designs around regulatory changes or technology grants to identify causal effects. - Cost-side studies estimating capital and recurring costs of ML adoption for Tier II banks to compute payback periods and inform subsidy policy.