From Sampling to Continuous Assurance: A Practical Framework for Using AI in Audit Transaction Testing

David Wang, DePaul University, Chicago, IL, USA

Santosh Vasudevan, Lead Data Scientist, Caterpillar, Chicago, IL, USA

 Executive Summary

Audit transaction testing has traditionally relied on sampling. Auditors review a subset of customer statements and compare key fields against the system of record. While this approach is well established, it leaves most transactions untested and becomes increasingly difficult to scale as organizations generate millions of customer documents each reporting cycle. Advances in AI-driven document intelligence now make full-population testing practical. Customer-facing documents can be converted into structured data and reconciled against authoritative records at scale, allowing auditors to focus on exceptions rather than routine verification. 
This article introduces the Risk-Based Population-Level Assurance Framework (RPLAF), a seven-stage, technology-agnostic approach for integrating AI-assisted evidence collection into audit planning, testing, exception management and continuous assurance under a confidence-aware, human-in-the-loop governance model. In a credit-card statement testing implementation, the framework expanded coverage from approximately 0.5% of the population to full-population testing, identified six confirmed discrepancies and reduced recurring testing costs by more than 94%. The framework is guided by a simple principle: AI gathers the evidence; auditors make the decisions. 

For a comprehensive analysis, detailed discussion, and supporting references, download the full article below