Literature on AI auditing reports large gains — roughly two-thirds higher efficiency and up to 70% better accuracy with 30–45% shorter audit times — but the improvements hinge on auditors' ethical readiness and human–algorithm collaboration.

Main Finding

AI-based auditing produces large, multidimensional efficiency gains relative to traditional auditing—combining faster processing, higher accuracy and lower costs—but these technical gains require complementary investments in auditor skills, governance, and ethical oversight. The systematic review reports average efficiency gains of ~68%, accuracy improvements of 40–70% (with some error-detection gains reported up to 92%), audit-time reductions of 30–45% (and specific public-sector gains of 8.6 seconds per case with 0.6% error), and cost savings in the range ~18–26%. However, sustained and legitimate efficiency depends on human–algorithm collaboration, interpretability, and institutional controls.

Key Points

• Scope and sample

  • Systematic Literature Review (PRISMA 2020) of Scopus articles published 2020–2025.
  • 323 initial hits → 30 peer‑reviewed articles met inclusion criteria (English, Scopus-indexed).
  • Geographic spread: Asia 47%, North America 28%, Europe 17%, multinational 8%.
  • Method mix: 60% quantitative, 25% mixed methods, 15% conceptual.

• Quantitative performance findings (synthesized)

  • Average efficiency increase: ~68%.
  • Accuracy improvements: commonly 40–70%; some studies report up to 92% in error detection.
  • Time reductions: typically 30–45%; example public-sector result: 8.6 seconds faster per case.
  • Cost savings: reported between ~18% and 26%.
  • Other: reported increases in reporting speed (~21%), reduction in human bias (~34%), and increases in consistency (~9%).

• Mechanisms and complementarities

  • AI automates data processing, predictive analytics and anomaly detection; auditors shift to roles as "cognitive supervisors" overseeing algorithmic outputs.
  • Success determinants: auditors’ digital literacy/training, organizational data readiness, governance & oversight, and system integrity.
  • Four organizational levers for internal audit success: commitment, access, capability, skilling.
  • Combined technologies (e.g., Blockchain + AI) amplify transparency and reduce error rates (example: 41% error reduction reported).

• Governance, ethics and accountability

  • Efficiency is framed as multidimensional: technical metrics (time/cost/accuracy) plus interpretability, transparency and moral responsibility.
  • Algorithmic oversight, ethical regulation, and open communication of model outputs are essential to legitimize AI-driven audit outcomes.
  • Firms with anticipatory ethical oversight saw higher audit reliability (~20% higher in some studies).

• Methods and theory used in reviewed literature

  • Common empirical methods: OLS regression, PLS-SEM, Random Forest, Gradient Boosted Decision Trees, neural networks.
  • Theoretical lenses: Agency Theory and Technology Acceptance Model (appearing in ~41% of papers), Diffusion of Innovation, UTAUT; others use ethical/professional theory.

• Limitations noted in the literature

  • Heterogeneous measures of “efficiency” across studies (operational vs. ethical constructs).
  • Selection bias: review limited to Scopus-indexed English articles (2020–2025).
  • Lack of long-run, causal firm-level evidence on labor-market and distributional effects.

Data & Methods

• Review design

  • Systematic Literature Review adhering to PRISMA 2020.
  • Database: Scopus; Boolean search: ("traditional audit" OR "manual auditing") AND ("artificial intelligence audit" OR "AI-based audit" OR "intelligent auditing") AND ("efficiency" OR "accuracy" OR "timeliness").
  • Inclusion: 2020–2025, peer‑reviewed journal articles, English, Scopus indexed.
  • Screening: removed duplicates, out-of-period items, non-relevant pieces; dual independent reviewers with consensus resolution.

• Data extraction & analysis

  • Concept-driven thematic coding and descriptive-comparative synthesis.
  • Quantitative synthesis via cross-tabulation matrix mapping AI application type (automation, predictive analytics, data mining) to indicators (time, cost, accuracy, transparency).
  • Quality appraisal adapted from Joanna Briggs Institute: clarity of objectives, method transparency, result–conclusion consistency (studies retained if meeting ≥2 criteria).
  • Triangulation performed between quantitative evidence and conceptual/theoretical findings.

• Empirical methods reported across studies

  • Statistical: OLS regression, PLS-SEM.
  • Machine learning: Random Forest, GBDT, Neural Networks.
  • Mixed methods: surveys, interviews, archival analysis for behavioral and organizational factors.

Implications for AI Economics

• Productivity and output measurement

  • AI auditing is a productivity-enhancing technology: large short-run gains in throughput, accuracy and cost per audit. Economists should account for multi-dimensional productivity (speed, accuracy, timeliness, and trust) rather than only time/cost metrics.
  • Standard productivity measures (TFP, value added per worker) may understate benefits if they ignore quality/accuracy externalities and reputational effects.

• Labor demand and skill-biased change

  • Evidence points to skill-biased technological change: demand shifts from routine verification tasks toward higher cognitive tasks (model supervision, interpretation, ethical judgment).
  • Expect occupational reallocation within auditing (fewer entry-level routine roles, greater demand for data-literate auditors, model auditors, and compliance experts).
  • Policy implications: need for reskilling programs, credential updates, and targeted subsidies for auditor upskilling.

• Complementarity vs substitution

  • Findings emphasize complementarity—human auditors remain essential for interpretability and accountability—so full substitution is unlikely in the near term.
  • The economic returns to AI investment will depend on complementary investments in human capital, data infrastructure, and governance mechanisms.

• Market structure and competition

  • Firms with better data readiness and governance capture disproportionate gains; this could increase concentration among large audit firms with resources to deploy and govern AI.
  • Regulatory standards and cross-firm interoperability could mitigate concentration risks.

• Public-sector and social-welfare effects

  • Faster, more accurate audits in the public sector improve fiscal oversight and service delivery; even small per-case time gains can scale to large efficiency dividends.
  • However, accountability and transparency concerns are central when audits affect public trust—social value of audit quality must be internalized in cost–benefit assessments.

• Regulatory and governance externalities

  • Algorithmic errors, opacity, and accountability gaps create negative externalities (misstatements, litigations, loss of trust). Effective regulation (algorithmic audits, reporting standards, model documentation) raises compliance costs but preserves social trust and legitimacy—these trade-offs should be incorporated into welfare analyses.

• Research gaps and suggested empirical strategies

  • Need for causal, firm‑level evidence on employment, wages, and productivity: use difference‑in‑differences exploiting staggered AI adoption, matched firm panels, or randomized rollout of AI tools.
  • Quantify the value of interpretability and accountability: incorporate reputation, litigation risk, and stakeholder trust into welfare models and cost‑benefit analyses.
  • Measure distributional effects across firm sizes, regions, and worker skill levels to inform labor and competition policy.

• Policy recommendations (economics-oriented)

  • Support training and certification programs for auditors in AI literacy and algorithmic oversight.
  • Encourage data‑infrastructure investments and standardized reporting to reduce adoption frictions and market concentration.
  • Create regulatory frameworks for algorithmic transparency, model validation, and liability allocation to internalize externalities and sustain public trust.
  • Fund longitudinal evaluations (randomized or quasi-experimental) to estimate net social returns of AI auditing, including non-pecuniary outcomes (trust, fairness).

Limitations of the review should caution economists: most evidence is short‑run and heterogeneous in measurement; conclusions about long‑run labor and distributional impacts require targeted causal studies.