Home Business & Finance
Current and Potential Applications of Text Mining in Accounting
Table of Contents:
Text mining has made its way to accounting, but its use among accounting professionals is only developing. While the “Big Four” are notable early adopters of these technologies, we are now seeing other accounting organizations starting to adopt or experiment with these technologies as well.
The current and potential applications of text mining technologies in accounting can be summarized in four families of applications: audit automation, accounting automation, tax automation, and business advisory applications.
Some accounting firms have reported using text mining software for automating contract reviews for audit or compliance assessment. Examples of adoption include Deloitte, EY, and EisnerAmper (Alarcon et al., 2019). Automated contract reviews can be performed as part of organizations’ annual financial audits and domain-specific audits such as fraud detection, internal audits, or IT audits. For example, contract reviews for assessing compliance with the latest lease accounting standards has been automated using contract review software such as Kira (“Deloitte Harnesses the Power of Kira for Lease Accounting Contract Review,” 2020).
In addition to contract reviews, current or potential applications of text mining in auditing include the review of SEC filings and other corporate communications materials, employment agreements, financing agreements, customer contracts, vendor contracts, and other documentation in support of transactions, financial statements, and disclosures.
In their 2018 article, professors Ting Sun and Miklos Vasarhelyi of Rutgers University stress the importance of textual data in auditing and advocate for the use of deep learning techniques by auditors. Sun and Vasarhelyi also analyze the usefulness of the information provided by various textual data in auditing. Their article provides a guide for auditors to implement deep learning, making the case that deep learning can support audit decision-making in all audit phases, including planning, internal control evaluation, substantive test, and completion (2018).
Accounting automation has started to incorporate intelligent information extraction (i.e., information extraction using machine learning approaches) in several areas. One of these areas is data entry automation. Indeed, accounting applications have started to embed or integrate information extraction, machine learning software combined with OCR technology and RPA solutions. These applications automatically record transactions (such as accounts payable invoices) and process them, or simply suggest journal entries based on the data extracted, a knowledge base, and patterns learned by the system from the prior transactions.
Similarly, applications of these technologies exist in employee expense reporting, where systems now can leverage OCR technology and intelligent information extraction from receipts to automate data entry and process expense reports.
Text mining applications can also be found in procurement to manage vendor contract compliance. In these cases, text mining can analyze vendor contracts to ensure compliance with policies or improve visibility into procurement patterns.
Additional current or potential uses of text mining technologies include:
Text Mining 65 agreements, and others) for maintaining compliance with accounting standards or accounting due diligence
H&R Block and K.PMG are examples of tax service providers who have reported using IBM Watson for tax preparation (H&R Block, 2017; “K.PMG and IBM,” n.d.). At H&R Block, tax professionals have worked alongside with an IBM Watson-powered application to help clients ensure that every deduction and tax credit is found (H&R Block, 2017).
KPMG has built an application with IBM Watson to help clients secure R&D Credits. With the application, users can upload thousands of documents and analyze structured and unstructured data at rapid speeds to help identify projects that are eligible for R&D Credits, using natural language processing to understand the context (Brown & Rainey, 2018).
Another example of the adoption of machine learning-powered applications for tax preparation using structured and unstructured data is Intuit Inc. Intuit provides an application called Tax Knowledge Engine (TKE) that helps TurboTax users streamline tax preparation. The system delivers answers tailored to each tax filer by intrinsically correlating and intertwining more than 80,000 pages of US tax requirements and instructions based on an individual’s unique financial situation. The system suggests what questions to ask, based on user data, and assists with computations. It includes a built-in explanation capability such that the engine can explain back the computations at any moment, for any tax concept involved (Wang, 2019).
Text mining, powered by machine learning and Al, is likely to profoundly affect tax automation. The use of these technologies will continue to progress to further automate tax preparation and tax planning tasks, with the emergence of question answering systems and potentially also virtual assistants that augment humans in addressing client-specific tax questions.
Text mining can also be used for tax audits or tax litigation analysis in taxation (income tax, sales tax, property tax, international taxation, and others). For example, text mining can analyze invoices, contracts, or other documentation for the proper tax classification of transactions or assets and taxation.
The tax authorities themselves are also expected to be significant users of these advanced technologies for tax examinations, tax fraud detection, tax avoidance or evasion, or tax policy from the perspectiveof governments or regulators. As an illustration, refer to the following statement from the Internal Revenue Service (IRS, n.d.):
IRS must take full advantage of technology to improve decisionmaking. Modem technologies continue to change the way organizations in the private and public sectors deliver their mission, products, and services. Government executives believe digital technologies are critical to improving financial services, such as revenue collection, audits, cash management and claims management. The IRS must respond to other changes (e.g., process robotics, blockchain and artificial intelligence) and integrate technologies that enable more efficient mission delivery. For instance, the IRS has applied data and analytics to refine identity theft detection models, filters and business rule sets designed to detect refund fraud and noncompliance. By continuously monitoring their performance, the IRS has ensured a cycle of improvement in detecting and preventing identity theft, (p. 19)