Dear IRS—Here’s Where You Should Spend Some of That $80 Billion

The IRS plans to use some of the $80 billion burning a hole in its pocket to become a “digital first” tax collector to improve customer service-type interactions with taxpayers. It also pledged, along with the Biden administration, not to increase audit rates for individuals making less than $400,000.

Enforcement efforts have been allotted $47.4 billion of the $80 billion total. So if the IRS is looking for more places to invest those funds, here are some ideas for how further digitizing can crack down on the estimated gross tax gap of $496 billion.

Data Mining for AI Use

The first step to any larger artificial intelligence project is going to be data mining. Technologies such as natural language processing and predictive modeling require large amounts of data to be effective—and the IRS just happens to have access to such large amounts of data.

The mining extracts insights from the existing data using machine learning algorithms and plain old fashioned statistical analysis. In other words, the IRS’ first step, which it almost certainly has undertaken in large part, is preparing the data it already has and reducing it to computationally useful patterns and trends.

Natural Language Processing

NLP refers to a field of AI that focuses on computers analyzing and understanding human language. NLP can be used to perform myriad analyses and is one of the underpinning technologies in the new wave of AI tools such as ChatGPT.

For tax enforcement purposes, NLP has an application examining tax forms to determine keywords and phrases that may correlate with tax evasion. This can run the gamut from the types of things an individual is deducting to a particular phrase or common use of a set of terms that may indicate a quasi-tax shelter is at play. NLP also can be turned loose on social media and reveal the public goings-on in a high earner’s life that may contradict how said earner presents themself for tax purposes.

NLP also can be used to automate customer service. From chatbots to virtual phone assistants, NLP can provide faster and more accurate information to taxpayers, creating a positive effect on compliance. With an average wait time during filing season of 13 minutes, and 19 minutes during post-filing season, the IRS help line is responsive—but not chatbot-level responsive.

This photo illustration shows the ChatGPT logo at an office in Washington, D.C., on March 15, 2023.

Photographer: Stefani Reynolds/AFP via Getty Images

Predictive Modeling

Modeling is all the rage in AI and technology circles. AI sophistication ranges from the word prediction algorithm on your cellphone’s keyboard to the most advanced AI large language models. They all basically operate as predictive engines. They predict the next word, number, or pixel, given a set of terms or images in a prompt.

Just as predictive modeling can create a picture of a horse given a prompt asking for one (provided it’s been shown pictures of other horses) it can analyze past known fraudulent returns to illustrate where future audits may be most fruitful. These sorts of projects have been undertaken abroad, with some promise. However, policymakers and coders must be careful not toinject underlying biases.

Predictive modeling also can be used to analyze the tax gap. The $496 billion figure is an estimate, but by analyzing taxpayer behavior, the IRS can further target compliance efforts to the portions of corporate and high-net-worth returns that are most likely to provide the most bang for the AI buck.

Factors such as sector, past tax history, occupation, income, and other data points can help form a model of the noncompliant entity or high-net-worth earner. Predictive modeling can then be used to evaluate the impact of a given compliance program to better allocate resources and ensure the program will achieve its end result.

Proceeding Ethically

A quote often attributed to Abraham Lincoln goes something like, “If you give me six hours to chop down a tree, I’ll spend the first four sharpening the axe.” In the AI and machine learning context, if you gave me $47.4 billion to build a system to enhance auditing, I would spend the first $20 billion ensuring it could be done ethically and with safeguards in place to avoid the kinds of disastrous pitfalls that have beset similar initiatives abroad.

There’s no question AI is the future of auditing and that it could vastly improve tax administration and compliance. However, the IRS must be able to ensure fairness for all taxpayers and limit knock-on effects.

This will require collaboration among technologists, policy experts, and taxpayer advocacy groups to ensure that the endeavor doesn’t jeopardize fairness for taxpayers as it proceeds. Assuming a thoughtful approach and a conscientious plan to advance under those guidelines, however, AI and machine learning can play a critical role in improving tax administration and ensuring all individuals pay their share—and $47.4 billion is a terrific start.

This is a regular column from tax and technology attorney Andrew Leahey, principal at Hunter Creek Consulting and a sales suppression expert. Look for Leahey’s column on Bloomberg Tax, and follow him on Mastodon at @andrew@esq.social.