Listed Nigerian banks that invest in and disclose AI assets—and that deploy chatbots—show higher returns on assets; larger banks also outperform peers. The association is robust in the sample but rests on a small, non-random set of banks and does not establish causality.

Main Finding

AI-structured investments are positively and significantly associated with bank efficiency measured by Return on Assets (ROA) for Nigerian deposit money banks (DMBs). Specifically, higher AI software book value, greater AI software expense disclosures, and stronger AI-enabled operational automation (chatbots) each predict higher ROA. Bank size is also positively associated with ROA.

Key Points

• Data and scope: Ten purposively selected listed Nigerian DMBs, annual observations 2015–2024.
• Main independent variables:
  • AI Software Book Value (AISBV) — year-end value of software assets.
  • AI Software Expenses Disclosed (AISED) — annual software maintenance/license costs.
  • AI-Enabled Operational Automation (AIEOA) — presence/intensity of chatbots/virtual assistants (e.g., Ziva, Leo).
  • Bank Size (BSIZE) — ln(total assets), included as a control.
• Outcome: ROA = Net income / Total assets (dependent variable).
• Descriptive snapshot: mean ROA = 2.4% (range −0.5% to 5.2%); AISBV mean = 12.4; AISED mean = 4.8; AIEOA mean = 4.1; BSIZE mean = 15.84.
• Econometric specification: ROA = β0 + β1AISBV + β2AISED + β3AIEOA + β4BSIZE + e (panel data, ex post facto design). The paper reports positive and statistically significant coefficients for AISBV, AISED, AIEOA, and BSIZE on ROA.
• Theoretical framing: Resource-Based View (AI as a VRIN intangible resource), Agency Theory (AI reduces information asymmetry/agency costs), and Diffusion of Innovation (staggered adoption and complexity effects).
• Policy/recommendation highlights from the authors:
  • Banks should increase investment in AI infrastructure and staff capacity building.
  • Regulators should design guidelines for AI adoption in banks.

Data & Methods

• Design: Ex post facto study using secondary financial statement disclosures and firm reports.
• Sample: 10 listed DMBs on the Nigerian Exchange Group selected purposively for consistent annual reporting and AI/digital technology investment disclosures.
• Period: 2015–2024 (annual panel).
• Variables and measures:
  • Dependent: ROA (net income/total assets).
  • AISBV: year-end software asset book value (capturing capitalized AI/software assets).
  • AISED: disclosed annual software expenses (maintenance/licenses).
  • AIEOA: index/presence measure for chatbot/automation deployment.
  • BSIZE: natural log of total assets (control).
• Analysis: Panel econometric model (specified above). The paper reports descriptive statistics and regression analysis (exact panel estimator not explicitly detailed in the excerpt; findings summarized as statistically significant positive effects).
• Limitations noted or implied:
  • Purposive sampling of 10 banks limits generalizability to all Nigerian banks.
  • Measurement issues: heterogeneity in disclosure practices and how AIEOA is quantified (index vs binary) may affect comparability.
  • Causality: ex post facto observational design cannot fully rule out reverse causality or omitted variable bias.

Implications for AI Economics

• Capitalization and accounting of AI matter: Positive associations for both software book value and software expenses imply that (a) treating AI as capital (book value) aligns with gains in asset productivity, and (b) continuing operating investments (licensing/maintenance) also support returns. Empirical work in AI economics should distinguish capitalized AI assets from operating AI spend.
• Productivity returns to AI in financial services: The findings support the view that AI investments can raise firm-level asset utilization (ROA), consistent with AI as a productivity-enhancing intangible. This strengthens the case for modeling AI as a form of intangible capital in growth and production-function analyses.
• Complementarity between scale and AI: Bank size correlates positively with ROA alongside AI variables, suggesting scale may complement AI investments (larger banks capture more returns from AI). Models of AI diffusion and returns should incorporate firm size and market structure heterogeneity.
• Measurement and disclosure externalities: Variation in disclosure practices can hide the true extent and effect of AI investment. Standardized reporting for AI capex and opex would improve empirical identification and cross-firm comparisons.
• Policy and regulation: Positive ROA effects imply social value from AI adoption in banking (efficiency, better risk management). Regulators should balance encouragement (standards, guidelines, upskilling) with safeguards (model risk, governance, consumer protection).
• Directions for research in AI economics:
  • Causal inference: use quasi-experimental designs (instrumental variables, difference-in-differences around adoption events) to address endogeneity.
  • Granular measurement: distinguish AI types (ML models, NLP/chatbots, expert systems), capital vs. operating AI spend, and implementation depth.
  • Labor and distributional impacts: examine how bank-level AI adoption affects employment, wages, and reallocation across banks.
  • Market-level effects: investigate competition, entry/exit dynamics, and pricing behavior as AI diffuses through financial intermediaries.
  • Cross-country and sectoral comparisons to assess how infrastructure, regulation, and human capital mediate AI returns.

Caveat: The summary reflects the paper’s reported associations; causal claims require stronger identification strategies than the ex post facto panel analysis used.