Firms in Livingstone reporting greater AI use—especially natural language processing—also report substantially more accurate financial reporting, but the small cross-sectional survey (n=60) cannot establish causation.

Main Finding

Adoption of AI technologies in accounting is strongly and positively associated with financial reporting accuracy among selected firms in Livingstone, Zambia. Survey evidence (n = 60) shows a high reported level of AI adoption (mean = 3.90) and a strong correlation with reporting accuracy (r = 0.700, p < 0.001). Regression models indicate AI adoption explains ~49.1% of variation in reported financial reporting accuracy (R² = 0.491, p < 0.001); when separate AI dimensions are included they collectively explain 58.4% of variation, with Natural Language Processing (NLP) the only statistically significant unique predictor.

Key Points

• Sample and setting: 60 accounting and finance professionals from selected firms in Livingstone, Zambia.
• Measured AI adoption via five dimensions: Machine Learning (ML), Robotic Process Automation (RPA), Natural Language Processing (NLP), automated financial data processing, and automated reconciliation & anomaly detection.
• Financial reporting accuracy measured by perceived reductions in errors, reconciliation accuracy, anomaly detection, consistency/reliability, and statement accuracy.
• Methods: positivist philosophy, deductive approach, cross‑sectional quantitative design, purposive + stratified sampling, structured questionnaires.
• Analyses: descriptive statistics, Pearson correlation (r = 0.700, p < 0.001), linear regression (AI → accuracy: R² = 0.491, p < 0.001); five-dimension model R² = 0.584 with NLP the significant unique predictor.
• Reported policy/practice recommendations: increase AI investment, continuous professional training, improve infrastructure, strengthen cybersecurity & data governance, establish AI governance frameworks, and maintain human oversight.
• Implicit caveats in study: small sample size, cross‑sectional and perception‑based measures, purposive sampling limits generalizability.

Data & Methods

• Design: quantitative cross‑sectional survey of accounting/finance professionals.
• Sampling: purposive and stratified; final n = 60 respondents.
• Variables and measurement:
  • Independent variable: AI adoption (composite and five dimensions: ML, RPA, NLP, automated data processing, automated reconciliation/anomaly detection). Likely measured on a Likert scale (reported overall mean = 3.90).
  • Dependent variable: Financial reporting accuracy (perceptual indicators: error reduction, reconciliation accuracy, anomaly detection, reliability/consistency, statement accuracy).
• Statistical procedures:
  • Descriptive statistics to assess level of adoption.
  • Pearson Product‑Moment Correlation to examine bivariate association (r reported).
  • Multiple regression to estimate effect and partition variance (R² reported for single and multi‑dimension models).
• Key quantitative results:
  • Correlation: r = 0.700, p < 0.001.
  • Regression (AI composite → accuracy): R² = 0.491, p < 0.001.
  • Regression (five AI dimensions jointly): R² = 0.584; NLP identified as the significant unique predictor.

Implications for AI Economics

• Firm-level information quality and market outcomes:
  • Large explanatory power (≈49% by composite AI; 58% by dimensions) suggests AI adoption materially improves the accuracy of reported financial information, which can reduce information asymmetry, improve investor confidence, and potentially lower firms’ cost of capital.
  • NLP’s unique predictive value implies that investments in document/text processing (disclosures, notes, unstructured data extraction) may yield outsized benefits for reporting accuracy relative to some other AI tools.
• Productivity and audit/monitoring economics:
  • Automation (RPA, ML) that reduces repetitive errors can lower monitoring and audit costs and enable redeployment of labor toward higher‑value tasks; however, the study’s perceptual measures mean quantifying cost savings requires further work.
• Adoption barriers and policy considerations in developing economies:
  • Findings reinforce the role of complementary investments (infrastructure, training, cybersecurity, governance). For policymakers in similar economies, targeted support—subsidies, training programs, standards for data governance—may raise adoption and the attendant public‑good benefits of more reliable financial information.
• Distributional and labor effects:
  • While improved accuracy supports efficiency, wider AI deployment may shift skill demands in accounting (greater demand for data/AI skills). Policymakers and firms should anticipate and manage workforce transitions.
• Research and evaluation priorities for AI economics:
  • Needed next steps include causal/longitudinal studies to assess persistence and directionality of effects, objective measures of reporting errors and audit adjustments to quantify economic gains, cost–benefit analyses of specific AI investments (e.g., NLP vs RPA), and wider samples to evaluate heterogeneity across firm size, sector, and country contexts.

Limitations to keep in mind when applying these implications: the study uses a small, purposive sample and perception‑based outcome measures from a single city; therefore, economic extrapolations should be treated as indicative and need confirmation with larger, objective, and longitudinal data.