AI in IRS audits could be a game-changer, but comes with pitfalls

The Internal Revenue Service plans to leverage artificial intelligence to identify patterns of noncompliance in large partnerships, while also ramping up its examinations of high-income taxpayers and big corporations. But the agency will need to tread carefully with how it uses AI amid growing questions about the reliability of the technology.

"This is entirely consistent with the stated agenda of the department since the Biden administration came in, which is to do more with the high-income taxpayers and also the corporations, and the large partnerships are a part of that," said Mark Everson, a former commissioner of the IRS and currently a vice chairman at Alliantgroup, a tax consulting firm. "It's been a vexing challenge for the service for years. The service is struggling and will continue to struggle to add and train personnel in the enforcement areas of their duties."

He pointed to recent reports from the Treasury Inspector General for Tax Administration about revenue officers not complying with the need to work through taxpayer representatives (see story) and court cases on the backdating of penalty approvals

irs-building-engraving.jpg
The Internal Revenue Service headquarters in Washington, D.C.
Samuel Corum/Bloomberg

"There's a lot going on within the workforce that is drawing scrutiny, and you've seen the service react to that," said Everson. "Commissioner Werfel stopped the unannounced visits. But there's just going to be more scrutiny on the taxpayer privacy side, and there's a fine line here. When you start to spend billions of dollars on the technology, and you add that capability targeted to this area, it will generate a reaction. If there are footfalls in this area, you're going to have a lot of attention to it if there are problems. And what the service has here is it's got the same set of challenges. ... It behooves the service to move extremely carefully here."

He sees political ramifications as well. "There's a political imperative, because the administration has done everything it can to demonize the big corporations and the high-income people, so there's an imperative to act quickly, perhaps in the runup to the election," said Everson. "But that just ups the stakes of all this. They've got to be extremely careful because it will be looked at very carefully. If there is a mistake, and there is a rule that's not a good rule, or if somebody is improperly targeted, there will be hell to pay because the people they're going after are resourced to fight."

He believes IRS Commissioner Daniel Werfel understands he will need to execute the new strategy appropriately given his prior experience at the White House Office of Management and Budget and in the private sector. The AI technology may be helpful with auditing large partnerships with their complex tax returns. 

"The technology is beneficial to tax administration, but it's got to be used carefully," said Everson. "When I was at the service 15 years ago, there were complex partnerships, and now they've gotten exponentially more complex. But the service was having challenges keeping up with them, and that was well before some of the layering had taken place. You've got to have the right protocols in place, both as to how you develop that technology, and then what's being done when there's an indicator that there's a problem with the taxpayer. Which personnel are looking at that? How are the decisions made to act upon that? In this day and age, big businesses and the wealthy are taking full advantage of technology, so it's understandable the government would seek to do the same thing. But the government is under a different level of scrutiny than the individual taxpayers."

A look at the pitfalls

Colin Walsh, a tax principal and practice leader of the tax advocacy and controversy services group in the specialty tax team at Top 10 Firm Baker Tilly, sees AI as a potential sea change in how the IRS selects particular tax returns for further examination, particularly when it comes to auditing large partnerships. 

"Partnerships are particularly ripe for artificial intelligence because the IRS in 2015, through the Bipartisan Budget Act, created the Centralized Partnership Audit Regime, which essentially makes it a lot easier for the IRS to make an assessment against a partnership," said Walsh. "Historically, the IRS under [the Tax Equity and Fiscal Responsibility Act], the old audit regime to the extent they wanted to make adjustments to a Form 1065, had to go through a really complicated process of notices and appeal rights before they could finally make an assessment against a partner. But CPAR, or the BBA, allows the IRS to use artificial intelligence, identify issues, make changes based upon those issues, and make an assessment right away through this concept of an imputed underpayment, so it's streamlined, both in terms of the data as well as how they make the assessment based upon the data."

There are some potential pitfalls, however, if the IRS is seen as targeting certain kinds of taxpayers unfairly just because they fall outside the normal patterns.

"Certainly, there are perfectly explainable statistical anomalies that are out there all the time," said Walsh. "You'd like to think that a reasonable IRS agent, if presented with the reason for the statistical anomaly, would accept that information and close the audit with it with a no-change letter."

He pointed out that the IRS has talked about artificial intelligence for a long time, but in a different context.

"The IRS has used artificial intelligence for years now as a way to identify tax returns that could be susceptible to identity theft," said Walsh. "The IRS will look at things like where the claim was filed and the amount of the claim. They're looking at patterns in the data to try to figure out if a return is subject to identity theft. Baker Tilly has seen scenarios where returns are flagged for identity theft that are perfectly legitimate. Of course, we're appreciative that the IRS is making those efforts in trying to seek out at any step, but artificial intelligence is not a perfect science, so there will be explainable anomalies."

AI could help the IRS with its workforce shortages as it tries to ramp up hiring to meet the demands of increased enforcement efforts. Initially the agency will be pursuing 1,600 millionaires who owe at least $250,000 in taxes, and 75 large partnerships with an average of over $10 billion in assets.

"The problem they've had is they haven't had an adequate workforce to go after that inventory," said Everson.  "It's not surprising that they would have inventories of matters that they want to look at. What they're doing is they're making an announcement to say, 'Hey, we're serious about this.' Part of that is designed to serve as a warning to others who may be tempted to engage in some overly aggressive activity to say, 'Hey, we're looking at this.'"

He sees it as both an outreach strategy and a political strategy to defend the IRS's budget and its appropriation for 2024, with elections approaching next year.

"They're trying to shore up the lines of defense in a series of areas," said Everson.

The IRS hopes to avoid further budget cuts as Congress fights over spending ahead of a possible government shutdown.

"They clearly are best served from a planning purpose by having a stable amount that they know they can work with over the coming three to five years or more," said Everson. "That's the best environment. But there's a point at which the organization can get jittery about whether it's going to have adequate funding. It's got to train people that it really needs in the enforcement area, as it's been doing in the service area. It's a greater challenge in the enforcement, though, because you've got a lot of protocols and technicalities that people need to learn."

"It's going to be hard to pick up the people that they need in that area," he continued. "They can do that, but that's time-consuming. But then those very same people need to take that inventory and they need to work down that inventory. If they're pressed to do it too quickly, there will be problems. That collides with the folks who are saying, 'Well, we'll give you all this money. Where are the results?' They need to get some stuff done just to demonstrate that the American taxpayers who contributed to the IRS are getting their money's worth."

For reprint and licensing requests for this article, click here.
Tax IRS Artificial intelligence Tax audits
MORE FROM ACCOUNTING TODAY