The US Treasury has made closing the tax gap a priority, with President Joe Biden’s fiscal 2022 budget boosting the IRS budget by about $1.2 billion, an increase of over 10% from previous levels. Former IRS Commissioner Charles Rettig has estimated that the tax gap could exceed $1 trillion on an annual basis. With this renewed focus on the tax gap, it’s time to consider modern approaches to identifying tax fraud and providing tools to auditors that empower them to identify fraud more efficiently and more quickly close cases.
Tax fraud takes many forms, such as cash underreporting and business income passed through as personal income. Furthermore, there are many ways to commit the different types of tax fraud. Some methods include point-of-sale “zapping” software, simply withholding cash sales, and SNAP/EBT trafficking. Other methods are being refined and created by tax cheats.
Each type of fraud leaves telltale indicators in the data. While humans aren’t well suited to analyzing large data, advanced analytics processes and machine learning are perfectly suited to identifying and highlighting patterns. This is crucial for effective audit selection and, more importantly, adapting to and identifying new and emerging trends that will stop new types of fraud from becoming commonplace.
An effective audit selection process that applies analytical techniques to all the data in tax filings makes the downstream process more efficient. Big data analytics pipelines leverage computing power and techniques to flag outliers, identify rule-based indicators, and engineer data features that are valuable to the auditors who receive the case work from the audit selection process. Auditors will be armed with intelligently selected records that are far more representative of cases likely to result in adverse findings.
Audit Selection Benefits From Modern Analytics
The audit selection process has remained largely unchanged for years and isn’t rapidly adaptable to new forms of fraud. Random selection and rule-based selection rely on known fraud methods and patterns.
Rapidly adapting to emerging fraud schemes is critical to combating the tax gap and reducing fraud. Due to the changing legal landscape, including the Wayfair decision, there must be new methods of identifying taxable sales to allow state and federal governments to enforce their own policies. This cross-state trade will require an unprecedented level of cooperation, new laws, and new and emerging technologies to assist enforcement agencies.
The data available to perform audit selection is massive. The sophisticated methods for accurate record selection are computationally intensive and not feasible outside of cloud-based systems. Client-owned or leased server systems are expensive to maintain and scale up during tax season.
Additionally, the expertise required to leverage the available algorithms and advanced probabilistic methods properly isn’t within the skill set of the average analyst or auditor. Large-scale, highly automated systems are perfect for performing the work of audit selection, while auditors are critical in performing the actual audit work. This synergy increases efficiency and compliance, and it closes cases faster and with less expense.
Emerging technologies that will make enforcement more efficient also will become available to fraudsters. Advanced methods can eliminate sales from the record at the point of sale, and bookkeepers can use sophisticated code and workbook macros to make fraud less evident and allow it to slip past audit selection tools.
Generative adversarial networks, for example, can produce deep fake images. It also can make matrices of false but realistic looking data that is repeatedly compared to real data until it reaches a threshold that passes the tests of the adversarial network, and it can be used to produce realistic tax information that is entirely made up butauthentic in appearance. This means enforcement agencies should act now to implement modern solutions to audit selection, audits, and enforcement before the fraudulent technologies become too ubiquitous.
Leveraging Big Data Analytics Processes
Creating a pipeline that implements supervised and unsupervised machine learning, statistical analysis, rule-based flags, and other advanced methods would reduce the impact on personnel and increase the accuracy of audit selection processes. Mathematical and algorithmic operations that take hours to complete on a server station could be completed in just minutes using advanced computing tools in the cloud.
The operations that can help segment business types and identify data patterns could perform a deeper analysis on all of the data rather than relying on rule-based flags and known statistical patterns. This operation casts a wider net, identifying smaller sole proprietorship that tend to make up the bulk of the sales tax withheld—up to 64%. According to the IRS tax gap study, this number is rising year over year.
Tax cheats employing modern methods may avoid detection for years, if not forever, and the unreported funds will become harder to recover as time passes due to the ex-post detection and recovery actions. The most effective audit selection method empowers auditors to close their cases faster by identifying the underpaid tax and recovering it quickly before the case has become stale.
An audit cycle that sometimes can last years leads to outdated evidence and a smaller overall recovery of tax dollars. Without drastically increasing the amount of personnel, a modern and highly automated approach leveraging big data analytics is the only way to effectively identify and recover unpaid tax.
Driving Voluntary Compliance
Sophisticated and determined fraudsters have had the benefit of underfunded tax authorities for years, resulting in a stubborn noncompliance problem perpetrated by individuals and businesses that haven’t experienced adequate enforcement. A modern approach would identify and recover funds being withheld from the public in the form of tax noncompliance.
The sophistication of the tools available require a large amount of computation and storage but will make auditors and analysts more effective, efficient, and productive. Technological advancement would not only stop existing and emerging forms of fraud, but it also would drive voluntary compliance as fraudsters see compliance efforts ramp up.
This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.
Steven Purcell is a data scientist and manager at G2Lytics LLC, researching big data analytics and machine learning applications to detect and prevent tax fraud.
We’d love to hear your smart, original take: Write for us.