Jimmy Bowles, CPA, CIA, CISA, is a senior audit manager at LBMC.

As we navigate an environment with rapidly changing technological advancements, the audit world will need to integrate artificial intelligence (AI) into its approach to enhance accuracy, efficiency and insight. AI will be used as another tool by auditors as it will increase the ability to automate data analysis, detect anomalies and provide predictive insights. To make sure that auditors can successfully implement AI into their process, there will need to be policies, procedures and rules set into place. The following steps will guide firms to a successful implementation of AI.

Originally published in the November/December 2023 issue of the Tennessee CPA Journal.

8 Steps to a Successful AI Implementation

Step 1: Define Objectives and Goals

It is crucial to clearly define the objectives and goals for implementation of AI use. Auditors will need to define policies on use of data with AI to ensure that data used follows all privacy laws and existing data corporate policies. Identify areas in your audit process where AI can be used to address problems like reducing manual effort, improving risk assessment or enhancing fraud detection.

Step 2: Assess Data Availability and Quality

Evaluate the availability and quality of your financial data. Ensure that the data is well-structured, accurate and accessible. Insights from AI algorithms are only as good as the input data. Garbage in, garbage out.

Step 3: Select the Right AI Tools

We must choose AI tools and technologies that align with our audit objectives. Common AI tools available include machine learning algorithms for anomaly detection, natural language processing (NLP) for textual data analysis and predictive analytics for trend forecasting. Auditors will need to work with data scientists and AI experts to ensure that they have the correct tools for audit requirements.

Step 4: Data Preprocessing and Cleansing

Data will need to be preprocessed to ensure meaningful results. Data cleansing is the process of detecting and correcting corrupt or inaccurate records from a data set. This involves removing duplicates, correcting errors and standardizing data formats.

Step 5: Integration With Audit Process

Auditors will need to identify points in the audit process where AI can add value such as data analysis, risk assessment and fraud detection. AI will help automate, accelerate and enhance the audit process by allowing auditors to obtain evidence over larger and more complex sets of data, as well as removing time-consuming tasks that auditors must complete. This will allow auditors to apply more valuable skills to other areas.

Step 6: Monitor and Refine

AI performance requires continuous monitoring. Audit teams will need to set up alerts for unusual activities and regularly review AI outputs. Outputs from AI are subject to AI hallucinations, which are defined as a confident response by an AI that does not seem to be justified by its training data. Auditors will need to verify results are true and supported by the input data.

Step 7: Enhance Auditor Skills

Usage of AI by audit teams is meant to supplement and not replace auditors. Training of audit teams is crucial to understanding the capabilities and limitations of AI audit tools. Auditors will need training in how to interpret insights, validate results and make informed decisions based on AI recommendations.

Step 8: Ethical Considerations and Transparency

Audit teams need to maintain transparency in AI usage. They need to ensure that AI results are explainable and free from biases. Auditors also need to adhere to guidelines and regulatory requirements to keep trust and credibility.

Conclusion

The integration of AI into financial audits is a tool that can empower organizations with unprecedented insights and efficiencies. Implementation of AI, following the step-by-step guide, will allow auditors to use AI as a strategic ally. There are limitations and issues as to what AI can accomplish. Auditors need to ensure that data uploaded to open-source AI tools, like ChatGPT and Bard, do not contain any intellectual property or personal data. Names, Social Security numbers, addresses and other identifying information need to be removed from data before uploading to AI. Understand who owns the data provided to AI. Firms will not want to send data to an AI platform that will sell or supply privileged information to anyone who is able to access the platform. AI cannot make decisions. Auditors cannot rely solely on results provided by AI. Human intervention will be required to interpret the information provided by AI. With the right training and monitoring, audits can become more accurate, efficient and valuable, benefiting both auditors and their organizations.