Artificial Intelligence to Help Resource-Strapped IRS More Efficiently Identify Tax Crimes

April 26, 2019

Tax criminals, beware: like the hosts in Westworld, the Internal Revenue Service (IRS) is becoming sentient! The IRS is now using artificial intelligence (AI) technology to more efficiently detect tax fraud, hidden assets, money laundering, identity theft and other noncompliance. In the past, instances of tax-related criminal activity had to be manually discovered by revenue agents, which could take days or weeks – a losing strategy for an agency that has seen nearly $1 billion in budget cuts since 2010. Now, thanks to AI, the pace at which the IRS can process data and conduct criminal investigations has advanced in line with 21st century necessity.

The IRS launched a new chapter of its AI program with the signing of a seven-year, $99 million deal with Palantir Technologies in late 2018. With an ever-shrinking budget and a loss of about 150 agents per year due to attrition, the IRS is deploying machine learning, natural language processing, and graph analytics to successfully identify instances of fraud and illegality before financial loss occurs.

The IRS plans to mine data from tax returns, property returns, bank reports and even social media accounts. The project will then employ algorithms and AI to identify patterns where taxpayer noncompliance might be present. In addition, the IRS will use natural language processing – technology that enables a computer to read and translate filings, as well as contemplate their meaning, which will help predict IRS success in its appeals process. Through language processing, machine learning is able to provide indicators of why appeals were either won or lost, and thereby, provide the probability of future success for cases with similar facts. In a cost-conscious environment, the IRS is under pressure to be smarter about how it selects cases that are most likely to conclude well.

In a webcast hosted by the American Bar Association, the IRS revealed comparative data for the use of this technology and fraud detection. As an illustration, AI technology is able to spot 296 suspected fraudulent returns claiming nearly $1.3 million, according to Benjamin Herndon, IRS Chief Analytics Officer. Previously, 84% of these returns had not been flagged during manual review by agents. The application of this technology means that the IRS can better uncover blind spots. As Todd Egaas, IRS Director of Technology, Operations and Investigative Services, reminded us during the webcast, “We are a small agency tasked with enforcing tax law across 250 million Americans.”

For more of the latest tax law and IRS updates, contact Aprio’s Tax Services team.