Teaching codeless machine learning to auditors

When the Institute of Chartered Accountants in England and Wales (ICAEW) released its white paper “Artificial Intelligence and the Future of Accountancy” the organization set the intention of building a greater understanding of AI’s practical use in accounting. Following through on that intention, the ICAEW published an educational resource that breaks down how auditors can utilize machine learning without possessing coding skills. 

The resource, titled “Codeless Machine Learning for Auditors” begins by briefly defining the term clustering, a process in which an algorithm sorts data points into groups without the feedback of humans. Then, an example demonstrates how clustering capabilities in the data visualization software Power BI can help auditors identify outliers in a dataset to detect fraud. 

The recommended workflow is comprised of two major phases: data exploration and graph creation. The dataset used contained anonymized credit card transactions. Data exploration in this example aggregated things like total and average balances and purchases. Exploration also involved filtering out rows in which the credit limit column was blank and creating a correlation plot. The plot revealed relationships between variables further explored in the graph creation phase. 

The first graph created — a scatter plot with account balance on the x-axis and purchases on the y-axis — had a “find clusters” feature, which identified clusters based on the volume of the balances. The second graph created was a clustering graph, which used a clustering algorithm that creates clusters in which each observation belongs to the cluster with the nearest mean. The last graph created identified outliers in the clusters. 

The resource concludes by recommending auditors use this outlier identification to further investigate instances for fraudulent activity. It notes that while programming languages offer more flexibility, Power BI and tools like it still provide powerful machine learning capabilities that take little time to utilize. 

Additional resources:
ICAEW data analytics community

Credit card dataset for clustering

The resource gives clear instructions for how auditors can execute a workflow in Power BI. While a few technical concepts are mentioned, a deep understanding of these concepts is not necessary.

The recommended software, Power BI, is not as flexible as actual programming languages. Additionally, the resource just walks through a handful of tasks that make up a workflow and does not comprehensively teach auditors all the ways in which they can utilize machine learning without coding.

The dataset mentioned in the resource can be accessed for free on Kaggle, so one can practice executing tasks from the workflow in Power BI with the same dataset. 

This article does not constitute legal or other professional advice and was written by students as part of the Duke Ethical Technology Practicum.