AI promises faster, more accurate audits through real-time analytics and standardization, but entrenched professional practice, algorithm aversion and skills gaps slow full uptake. The result is not wholesale replacement but a continuous rebalancing of tasks—routine checks are automated while judgment, oversight and new practices rise—requiring firms to redesign work and training to realize benefits.

Main Finding

AI is reshaping auditing by enabling efficiency, accuracy, continuous/real‑time examination of full data populations and new analytical capabilities, but its full adoption is constrained by sociotechnical barriers (resistance, algorithm aversion, transparency/explainability, skill gaps, complexity). From a sociomaterial perspective, the auditor’s role is being continuously reconfigured into a shared human–AI agency: auditors move toward oversight, interpretation, and governance of AI systems while AI assumes performative operational roles.

Key Points

• Drivers of AI adoption in auditing

  • Efficiency gains (automation of routine tests, faster analysis)
  • Improved accuracy and anomaly detection (machine learning, DL)
  • Real‑time / continuous auditing (examining entire populations vs. sampling)
  • Big Data analytics and NLP/LLMs for unstructured documents
  • Standardization and scalability of certain audit procedures
  • Significant investments by large firms and startups (Big Four initiatives; GenAI uptake in practice)

• Barriers to adoption

  • Resistance to change and cultural inertia in firms
  • Algorithm aversion, heuristics and cognitive biases among auditors
  • Lack of transparency / explainability of complex models (black‑box concerns)
  • Expertise and training gaps (analytical, data, AI governance skills)
  • Technical complexity and integration challenges with legacy systems
  • Governance, ethical and regulatory uncertainty

• How AI changes audit work (mapped to audit phases)

  • Engagement acceptance: AI can screen external data and estimate risk profiles
  • Planning: NLP and pattern detection help identify risk areas and map processes
  • Risk response: continuous control monitoring, population‑level substantive procedures
  • Completion/reporting: potential for continuous or scaled opinions (vs. categorical)

• Theoretical framing (sociomateriality)

  • Uses sociomateriality / imbrication: material agency (technological capabilities) + social agency (human responses)
  • AI introduces forms of “artificial agency” (autonomous, learning agents) that co‑constitute practices with auditors
  • Reconfiguration is dynamic and context dependent — technology and practices recursively shape each other

• Practical recommendations (from reviewed studies)

  • Invest in human capital and training (data, AI governance, interpretive skills)
  • Develop governance frameworks, mandatory human review and explainability standards
  • Design organizational practices that facilitate continuous human–AI interaction and oversight

Data & Methods

• Study type: Systematic Literature Review (SLR)
• Methodological protocol: PRISMA; approach based on Tranfield et al. (2003)
• Databases: Scopus and Web of Science
• Search string: audit* AND ("artificial intelligence" OR "natural language processing" OR "machine learning" OR "deep learning" OR "artificial neural network")
• Selection: initial 1,286 records (Scopus 650; WoS 636); 238 duplicates removed → 1,048 unique records screened; final sample analyzed = 43 studies
• Analytical lens: sociomateriality theory (Orlikowski; Leonardi; Barad), focusing on interaction of material agency (AI capabilities) and social agency (auditors, organizations)

Implications for AI Economics

• Labor and skill composition

  • Skill‑biased technical change: demand shifts from routine audit tasks to data/AI oversight, analytics, and judgment roles; upward pressure on wages for AI‑capable auditors
  • Potential reduction in demand for low‑skill audit labor but complementary increase in high‑skill roles (retraining and transition costs important)

• Productivity, pricing and market structure

  • Productivity gains in audit production (faster, broader coverage) could lower marginal costs and alter fee structures, but realized benefits depend on firms’ ability to integrate AI
  • Large firms with capital to invest (Big Four) gain competitive advantages, potentially increasing market concentration; startups and niche tools can also shape competition

• Audit quality and information economics

  • Continuous auditing and full‑population testing can reduce information asymmetries and increase the timeliness and granularity of assurance, with potential spillovers to capital markets
  • However, opacity and biases in AI models create new sources of information risk; regulators and users may discount algorithm‑driven assurances absent transparency

• Adoption dynamics and externalities

  • Training and governance are public‑good–like investments: underinvestment by individual firms can slow diffusion; policy or industry coordination may be needed
  • Network effects and standards (data formats, explainability norms) will influence speed and direction of adoption

• Regulatory and welfare considerations

  • Policymakers should weigh benefits of improved audit coverage against risks from opaque models, accountability gaps, and concentration
  • Regulation that mandates explainability, validation, and human oversight can influence firms’ cost of adoption and the competitive landscape

• Research needs (economic empirical agenda)

  • Firm‑level causal evidence on how AI adoption affects audit quality, auditor labor demand/wages, and audit fees
  • Longitudinal studies of reallocation effects within audit firms and across markets
  • Measurement of productivity gains vs. integration and governance costs
  • Analysis of market concentration dynamics and entry by AI‑driven providers
  • Evaluation of regulatory interventions (mandates on explainability, validation, liability) on adoption and outcomes

Short takeaway: AI has strong productivity and informational potential for auditing but realizing those gains depends critically on sociotechnical investments (skills, governance, explainability). For economists, the main open questions concern the size and distribution of productivity gains, labor reallocation, market structure effects, and the welfare implications of AI‑mediated assurance under different regulatory regimes.