Banks that talk openly about AI attract more customer deposits in India, with public sector lenders benefitting most; private banks show a credibility gap from weaker disclosure. The study interprets voluntary AI reporting as a signaling mechanism that bolsters depositor trust, though evidence is correlational and based on a small sample of listed banks.

Main Finding

Voluntary AI disclosure in banks’ annual reports is positively associated with customer deposit growth. Using a balanced panel of 12 Indian banks (Nifty Bank constituents) over 2019–2023, a constructed AI disclosure index has a statistically significant positive coefficient in a random‑effects log(deposits) regression: a 10‑point increase in the AI index is associated with about a 2.5% increase in deposits (coef = 0.002514, p < 0.001). Ownership matters: private banks (dummy = 1) have substantially lower deposits than public banks (coef = −2.015, p < 0.001). Net profit is positively associated with deposits; BVPS and the policy repo rate are not significant.

Key Points

• Sample and scope: 12 banks listed on the Nifty Bank Index (9 private, 3 public), annual reports 2019–2023 (60 observations, balanced panel).
• AI disclosure measure: AI Disclosure Index built from keyword frequency in annual reports (three categories: digital awareness/transformation; AI applications/products/processes; AI-related challenges & cybersecurity). Base year 2019 = 100. Content analysis done with MAXQDA24.
• Main econometric model: Panel EGLS with cross‑section random effects; Hausman test (χ2 ≈ 0, p = 1.000) supports random effects over fixed effects.
• Key regression results (dependent = log(deposits)):
  • AI INDEX: coef = 0.002514, SE = 0.000409, p < 0.001 → ~2.514% higher deposits per 10‑point index rise.
  • DUMMY (private = 1): coef = −2.015, SE = 0.480, p < 0.001 → private banks have much lower log(deposits) versus public banks (exp(−2.015) ≈ 0.13).
  • NET PROFIT: positive and significant (small marginal effect per unit), p ≈ 0.022.
  • BVPS and PRR: statistically insignificant (p > 0.1).
• Model fit: R² ≈ 0.62, adjusted R² ≈ 0.586.
• Diagnostics: Deposits log‑transformed for stationarity; standard tests for heteroskedasticity/autocorrelation/multicollinearity reportedly done (details not fully reported).

Data & Methods

• Data:
  • Unit of analysis: bank-year (12 banks × 5 years = 60 observations).
  • Dependent variable: customer deposits (log transformed).
  • Explanatory/control variables: AI Disclosure Index, Net Profit (NP), Book Value Per Share (BVPS), Policy Repo Rate (PRR), ownership dummy (private=1, public=0).
  • Descriptives: mean deposits ≈ INR 804,468; AI index mean ≈ 165.67; NP mean ≈ 9,395.5; PRR mean = 5.03.
• AI Disclosure Index construction:
  • Keyword list informed by FSB, OECD, IOSCO guidance; grouped into 3 categories.
  • Annual reports coded with MAXQDA24; keyword frequencies aggregated and indexed (2019 = 100).
• Estimation:
  • Random effects panel EGLS chosen after Hausman test.
  • ADF tests used for stationarity; log transformation applied to deposits.
  • Reported robustness/diagnostics mentioned but detailed tests/results (e.g., heteroskedasticity corrections, clustering, alternative specifications) are limited in the paper.

Implications for AI Economics

• Signaling value of tech disclosure: Voluntary reporting on AI initiatives appears to function as a credible signal that reduces information asymmetry and strengthens depositor confidence, with measurable effects on funding inflows (deposits). This highlights that non‑financial tech disclosure can have real financial consequences.
• Ownership moderates disclosure effects: Public banks retain a trust/credibility advantage; private banks’ lower baseline deposits and the paper’s interpretation of a “credibility gap” suggest heterogeneous returns to disclosure across ownership types. Regulators and managers should consider ownership when designing disclosure policies and communication strategies.
• Incentives for disclosure and adoption: If AI disclosure raises deposit inflows, banks have an additional private incentive to invest in and communicate about AI—potentially accelerating diffusion. However, disclosure may reward communicative ability as much as actual operational adoption (reporting intensity vs. implementation).
• Policy and regulatory design: Findings support arguments for standardized AI reporting frameworks (taxonomy and disclosure guidelines) to improve comparability and credibility—reducing the chance that keyword‑based disclosures merely reflect marketing rather than substance.
• Research directions for AI economics:
  • Causal identification: address potential endogeneity (e.g., better‑performing banks both adopt AI and attract deposits) using instruments, lag structures, or quasi‑experimental designs.
  • Richer measures: combine disclosure indices with observable measures of AI deployment (capex, product rollouts, third‑party audits) and customer survey/behavioral data to separate signaling from real service improvements.
  • Broader financial effects: examine impacts on cost of funds, lending spreads, market valuations, and competitive dynamics across banking sectors and countries.
  • Distributional and welfare angles: investigate whether AI disclosure changes who banks attract (retail vs. corporate), alters competition for deposits, or affects financial inclusion.

Limitations to note for interpretation: small sample (12 banks), short time span (5 years), index based on keyword frequency (may capture reporting style rather than substantive AI use), and potential reverse causality or omitted variable bias not fully ruled out.