From random picks to targeted insights: AI’s role in better audit sampling.
Sampling is a fundamental element of financial auditing. Traditionally, auditors rely on sampling techniques to test and verify large volumes of financial transactions or records without having to examine each one individually. According to ISA 530, sampling involves selecting items from a population to gather enough evidence to make reliable conclusions. This process typically involves “random” or “systematic” selection, which is meant to produce a representative sample of the whole population.
While effective, traditional sampling methods come with limitations. Selecting samples manually or through automated, yet random, methods can miss out on subtle anomalies that might signal potential issues or risks. This can be particularly limiting in complex environments with vast, varied transaction data, where certain unusual patterns or potential indicators of fraud may not always be evident. It can also be difficult to select a truly random method of selecting samples and therefore human bias can creep in to sample selection.
Artificial Intelligence (AI) offers a promising alternative to traditional sampling methods by analyzing data more dynamically and intelligently. For instance, an AI model trained to analyze financial data could quickly digest entire datasets, such as a debtors ledgers or purchase nominal transaction histories. Instead of merely picking a random sample, AI could identify transactions that fit particular criteria and select samples based on patterns that stand out.
If we were to give an AI model specific criteria to flag certain behaviors, it could prioritize transactions with potential fraud indicators or irregular patterns. This means AI would not only provide a random sample but would also ensure the selection includes data points that might indicate a risk. In essence, it shifts sampling from a “random pick” to a more “risk-focused” approach. By using AI in sample selection, auditors could increase the likelihood of catching anomalies that could otherwise be missed with traditional sampling methods.
So how could this work in practice?
Imagine you’re auditing a company’s purchase transactions for the year. Traditionally, you might randomly select a sample of these transactions to review, based on a stratification, hoping that your sample represents the overall dataset. However, with AI, you could input the entire purchase nominal history, materiality figures and ask AI to calculate the sample size, perform the stratification calculation, and then select a random sample and To target transactions with a higher likelihood of containing issues, we’d provide the AI with certain criteria to look for. Here’s what these criteria could include:
- Unusual Vendors: Flag purchases from new or rarely used vendors, as these might carry more risk.
- Large Amounts: Select transactions above performance materiality
- Frequent Split Transactions: Identify instances where a large purchase appears to be split across several smaller transactions to avoid review thresholds.
- Outlier Patterns: Look for transactions with unusual characteristics compared to the average transaction, such as an unusually high price-per-unit or large quantities purchased.
An AI bot could be trained for this purpose to make the sample selection process more efficient and insightful. It could also then create a report explaining how the sample selection has been performed to go on file.
This type of AI-enhanced sampling offers several benefits beyond the traditional “eenie meenie minie mo” approach. For auditors, AI can provide a more efficient and effective method of reviewing transactions by focusing their efforts on high-risk items. Instead of sifting through a large dataset with a randomly selected sample, AI allows auditors to concentrate on areas of greatest interest or concern, potentially leading to a faster, more thorough audit.
For clients, AI-powered sampling can mean a more accurate audit with fewer interruptions. Since AI can highlight potentially risky transactions right from the start, auditors are less likely to need repeated queries, making the process smoother for everyone involved. And for stakeholders or readers of financial statements, an AI-assisted audit can provide greater confidence in the audit’s thoroughness and reliability, knowing that advanced tools were used to identify and address possible anomalies.
Traditional sampling methods, while effective to an extent, come with limitations that AI could help overcome. With its ability to process large datasets quickly and identify unusual patterns, AI presents an opportunity for auditors to improve the precision and reliability of sample selections.
However, the technology still has some limitations. Current AI models can struggle with large datasets due to token limits, which restrict the amount of data they can handle at once. There’s also the issue of AI hallucinations, where the model might create phantom data points, leading to inaccurate insights. Additionally, applying complex criteria across a large dataset can sometimes overwhelm current AI systems.
With advancements like ChatGPT o1, many of these issues may be addressed, thanks to enhanced reasoning capabilities and larger capacity. As AI technology continues to evolve, the potential for a smarter, more targeted approach to audit sampling will only increase, benefiting auditors, clients, and financial statement users alike. The future of audit sampling may indeed be more intelligent, efficient, and secure with AI in the mix
AI has the potential to bring a more refined, risk-based approach to audit sampling, offering greater efficiency and accuracy than traditional methods. While challenges remain, these will likely decrease as AI continues to advance. For auditors, the promise of AI lies not only in simplifying their work but also in ensuring the integrity of the financial data they assess.
