Audit Data Analytics model - from Sceptical to Suspicious

6,000

Traditional auditing approaches rely on a sceptical mindset that questions data integrity but often lacks actionable direction, limiting effective fraud detection. This SURF advocates adopting a suspicious mindset, where auditors generate and test precise, causally grounded hypotheses. We propose a novel causal Audit Data Analytics (ADA) framework that integrates do-calculus and Bayesian reasoning to improve fraud detection in hotel revenue auditing. Do-calculus enables unbiased estimation of causal effects from observational data by identifying valid adjustment sets via manipulation of causal graphical models, addressing confounding and selection biases. Bayesian inference supports dynamic updating of fraud hypotheses as new evidence emerges, formalizing suspicion into probabilistic terms. Applying directed acyclic graph (DAG) modelling to capture causal relationships among pricing strategies, seasonal demand, market competition, and revenue, we demonstrate how do-calculus identifies adjustment variables, enabling interpretable causal effect estimation. Coupled with Bayesian updating, this framework empowers auditors to detect deviations indicative of fraud. Implemented as a scalable RESTful API, the framework integrates seamlessly with existing audit workflows. Experimental evaluation on simulated hotel revenue data confirms superior accuracy in causal effect estimation and fraud detection compared to correlational baselines. This work bridges causal inference theory and auditing practice, supporting auditors in transitioning from generalized skepticism to targeted suspicion, thereby enhancing audit effectiveness and decision-making in a data-driven era.

The expected outcome is an Audit Data Analytics DAG construct with an API as products.