By Bruce E. Lane, CPA
In my role as a business applications technical specialist at Microsoft Corp., I work with clients to determine the business application fit for their organization. That fit is evaluated both from a technical and a business process perspective. The problem, however, is that clients often develop business processes to conform with the technical limitations of the current enterprise resource planning (ERP) system. Therefore, we need to be careful not to fall into the trap of replacing an incumbent system with a new system equivalent rather than a system that is more efficient and effective.
This problem can be largely avoided if we undertake a “digital transformation.” A digital transformation entails a holistic view of desired outcomes that can be supported by technology. Often, a business’s current process must be reimagined since it may be dependent on outdated technology.
A digital transformation is a necessary step for organizations looking to artificial intelligence (AI) to improve ERP processes. AI provides insights that were not possible in our previous systems. AI is not an out of the box solution that immediately provides valuable insights; it must be trained over and over with positive reinforcement until it reaches an acceptable accuracy rate. This is only possible with large volumes of accessible data that AI models can access.
Here, I will discuss a feature that I also demonstrated at the PICPA Accounting & Auditing Conference in December 2020: the customer payment predictor.
When auditing accounting estimates such as accounts receivable, we often rely on history and trends to support the fairness of the client’s estimate for collection. The collection rates of prior years are a common way to determine the fairness of the allowance for doubtful accounts in the current year. This is typically a macro-level exercise, and, in some instances, the auditor will examine individual documents.
AI tools can make this task more efficient by providing the ability to predict collection dates with little to no effort. An AI system could provide a detailed list of predicted collection dates of the entire population of accounts receivable documents. Auditors can use this list to determine the fairness of the accounts receivable presentation on the financial statements.
The auditor needs to plan the audit procedures based on his or her experience and the assessment of related risks. I am not about to suggest exactly how an AI application should be audited, but rather I will discuss and present how the application makes its predictions.
Here is a basic definition of a customer payment predictor: A machine learning model that leverages historical invoices, payments, and customer data to predict when a customer will pay an outstanding invoice.
The prediction is based on the model processing historical data, identifying relationships, and then projecting how current data will resolve based on that historical data. The model can be continuously trained to improve results once exposed to its prediction so that it can process its inaccuracies to further identify relationships and the cause and the effect to improve results.
Columns S through AC in the image below are sample invoice files. Note column AA, which the AI application uses to train itself. This particular AI application uses three buckets: on time, late, and very late.
The AI parameters setup window is used to enable the AI capability and classify what entails “very late.” It also presents the current accuracy and allows you to set the schedule for processing predictions and modify the data variables used to train the model. For instance, the user may add location, shipping method, or item detail to improve the model if it is deemed those variables influence the timeliness of collection.
The detailed prediction window presents the results of the AI prediction. Note that highlighting an individual record will present more detail in the window on the right side.
More information on the customer payment predictor can be found at Microsoft.com.
AI in ERP systems is just beginning to grow, and it will take time for the technology to mature and for accounting personnel to accept it as a reliable tool. Understanding and accepting predictive analytics, however, will provide us all with tools to be more efficient and effective.
Bruce E. Lane, CPA, is a technical specialist U.S. – Dynamics 365 finance & operations/commerce – for Microsoft in the Pittsburgh area. He can be reached at email@example.com.
Sign up for weekly professional and technical updates from PICPA's blogs, podcasts, and discussion board topics by completing this form.