The IRS is actively integrating artificial intelligence (AI) and machine learning into its audit selection process to identify high-risk returns more effectively. While the agency has used statistical models for decades, AI represents a significant advancement, moving beyond simple red flags to analyze complex patterns and relationships in vast datasets.
This new approach, supported by funding from the Inflation Reduction Act, is intended to close the “tax gap”—the difference between taxes owed and taxes paid.
How AI is changing the audit selection process
- Refined analysis: AI models analyze massive datasets, including returns and third-party information from various financial institutions and digital platforms. This creates a more sophisticated and precise risk assessment than previous manual or less advanced statistical methods.
- Predictive risk scoring: Machine learning algorithms assign a risk score to tax returns, allowing the IRS to prioritize cases with the highest likelihood of non-compliance. This enables more targeted and efficient audits.
- Greater data cross-referencing: AI can cross-match data from numerous sources, including W-2s, 1099s, K-1s, cryptocurrency exchanges, and real estate databases, to identify discrepancies faster and more accurately.
- Focus on high-risk sectors: Initial AI-powered audit programs are primarily focused on complex, high-value cases that are historically difficult to audit effectively. This includes large partnerships, high-income individuals, and corporations.
- Continuous learning: The AI systems are designed to learn and adapt over time, refining their ability to identify new patterns of tax avoidance and emerging fraud schemes.
Who is most likely to be flagged by AI?
While the IRS has stated it will not increase audit rates for those earning less than $400,000, AI’s targeting can still impact everyday taxpayers, freelancers, and small business owners. Common issues that AI may flag include:
- Income discrepancies: Mismatches between what a taxpayer reports and third-party data from employers, financial institutions, and digital platforms.
- Underreported income: Taxpayers, including gig workers and freelancers, who report low income relative to indicators of wealth or industry norms.
- Unusual deductions: Large charitable deductions that appear disproportionate to income, or round-number expenses that suggest estimates rather than documented figures.
- Digital asset transactions: Cryptocurrency transactions not properly reported.
- Complex business structures: Multi-tiered partnerships and business returns with extreme or inconsistent ratios.
- Multi-year inconsistencies: Financial data that does not reconcile from one year to the next.
Key benefits and concerns
Benefits:
- Increased efficiency: AI helps the IRS focus limited resources on the most likely instances of non-compliance, avoiding “no-change” audits that waste time and burden compliant taxpayers.
- Enhanced fairness: Proponents argue that AI can reduce human subjectivity, leading to more consistent and targeted enforcement.
Concerns:
- Lack of transparency: The IRS is very private about its algorithms and the data used for training them, which makes it difficult to verify their fairness or accuracy.
- Potential for bias: Some studies have raised concerns about potential algorithmic bias, such as a 2023 report showing higher audit rates for Black taxpayers claiming the Earned Income Tax Credit (EITC).
- Impact on compliant filers: AI can flag taxpayers for simple mistakes, subjecting them to greater scrutiny even without intent to evade taxes.
- Need for human oversight: Experts emphasize the importance of retaining human oversight in the process to prevent flawed algorithms from causing harm.