By Thomas J. Healey
Conservatives determined to defund the IRS should reexamine their stand on the basis of one inescapable fact: every dollar of unchecked tax fraud is a dollar more on the shoulders of taxpayers who don’t try to cheat the government. In 2021, the U.S. Treasury estimated that the “tax gap” — the difference between taxes owed and those collected — was almost $600 billion.
The good news is that artificial intelligence (AI) now holds the promise of transformative change for this country’s sprawling and antiquated tax system. A portion of the Inflation Reduction Act signed into law by President Biden in August 2022 plows some $80 billion in funding over the next 10 years into the IRS for upgrading computer systems, strengthening enforcement, and bolstering its staff, which dwindled from nearly 95,000 in 2010 to around 79,000.
Even with the lower IRS funding in the debt limit package passed by the Senate Thursday night, there is still more than adequate money to take full advantage of the power of AI.
The potential benefits to the nation’s tax agency have never been more apparent. Consider the fact that each filing error by taxpayers creates an extra step for IRS personnel, resulting in enormous backlogs when compounded millions of times over. As of 2020, there were 20 million unexamined tax returns.
An AI algorithm, however, could be “trained” on the agency’s massive data sets to pick up on the most commonly made mistakes, then proactively reach out to taxpayers with information and instructions on how to correctly fill out the most error-prone IRS forms.
That’s just the tip of the AI iceberg. By automating routine tasks, identifying patterns of non-compliance and fraud, and providing better customer service, the IRS might finally be able to propel itself into the 21st century.
Many European governments have already, adopted sophisticated fraud-detection technologies with impressive results. In Denmark, for example, tax officials now identify an estimated 60% of fraud cases through machine learning applications.
In the critical field of predictive modeling, automation could allow the IRS to forecast the likelihood of non-compliance or nonpayment by individual taxpayers by analyzing tax returns, financial statements, and other historical data. Problematic taxpayers could then be assigned risk scores, allowing tax authorities to take steps to encourage compliance, rather than react after a problem has occurred.
Through continuous data mining, AI tools could also help to identify broad patterns of behavior that indicate the potential for non-compliance or other irregularities and prioritize individuals who fit these templates for audits or other enforcement action. Likewise, being able to predict which taxpayers are likely to be delinquent in their tax payments could help the IRS allocate resources needed for collection and improve the overall collection rate.
IRS special agents in particular could benefit from these automated capabilities. Currently, they must look for hidden assets, money laundering, identity theft and much more with the naked eye. As experience has shown, this can be extremely time-consuming and vulnerable to errors. The audit rate of individual income tax refunds fell from 0.90% in 2010 to 0.25% in 2019.
In a webcast hosted by the American Bar Association, IRS officials revealed that research and investigative techniques that used to take weeks or months could be done in minutes with new technology the agency said it plans to roll out to detect taxpayer noncompliance.
In terms of customer service, automation could also portend desperately needed change. An AI-powered phone response system built on natural language processing could allow the IRS to respond to the most common taxpayer inquiries, freeing agents to resolve more complex issues and, in the process, address the fact that only 11% of the 282 million phone calls from taxpayers seeking help or guidance during the 2022 filing season actually reached an agent.
Taxpayers seeking to reach customer service agents on the telephone are waiting on hold for an average of 29 minutes. Through predictive dialing, AI could also analyze past interactions with taxpayers in order to determine the best times and methods to reach them, increasing the chance of a successful contact and thus improving collections.
Everyone, even Republicans, should be loudly applauding this landmark $80 billion commitment to strengthen, over the long run, the IRS, knowing that enhanced enforcement alone would pay for the investment many times over.
Thomas Healey is a senior fellow at Harvard University’s Kennedy School of Government. He was an assistant secretary of the U.S. Treasury under President Ronald Reagan.
Here’s how to submit an op-ed or Letter to the Editor. Bookmark NJ.com/Opinion. Follow us on Twitter @NJ_Opinion and on Facebook at NJ.com Opinion. Get the latest news updates right in your inbox. Subscribe to NJ.com’s newsletters.
Follow us on Twitter @NJ_Opinion and on Facebook at NJ.com Opinion.