Does Artificial Intelligence Improve Audit? Reconciling Divergent Empirical Evidence on Audit Quality, Efficiency, and the Audit Workforce
0 viewsAbstract
Audit firms have spent more than a decade investing in artificial intelligence (AI), yet the academic evidence on whether those investments improve the audit remains openly contradictory. This study asks why such a divergence exists and what the accumulated evidence, read as a whole, actually supports. This study conducts a structured evidence synthesis of seventeen empirical studies of AI and audit data analytics in external auditing published between 2015 and 2026, coding each study on five dimensions: evidence type, unit of analysis, technology construct, outcome domain, and direction of the reported effect. Vote-counting is then performed within strata rather than across the full corpus, so that studies measuring different things are never pooled. Three findings emerge. First, every one of the seven studies that measures a realized audit outcome — restatements, going-concern accuracy, misstatement detection, inventory count error, audit fees, or task completion time — reports a favorable effect, with magnitudes ranging from a 5.0% reduction in restatement likelihood to a fall in inventory count time from 681 hours to 19 hours. Second, all three studies that measure human judgment report adverse or mixed effects, because auditors discount machine-generated evidence and reviewers penalize procedures that appear less effortful. Third, the workforce divergence is not a genuine empirical contradiction but an artefact of the unit of analysis and of the outcome definition: firm-level headcount of accounting staff and office-level counts of auditor positions are different quantities that can legitimately move in opposite directions.
Keywords
Brand Credibility, Brand Equity, Brand Trust, Branding, Consumer Demand, Price Sensitivity, Willingness To Pay