Former IRS Commissioner Explains How AI Was Used to Deliver Immediate Value Amidst Taxpayer Scrutiny of Spending
As businesses continue to allocate substantial funds to AI, there’s a growing expectation from both employees and shareholders to see concrete results. However, by the close of 2025, a mere 15% of executives anticipate that AI implementations will boost profits. To understand how companies can bridge the gap between AI enthusiasm and actual returns, we can look to an unexpected source outside the private sector: the Internal Revenue Service.
While a government agency may not seem like a typical business, it faces the same pressure to demonstrate AI progress and the same difficulties in achieving it. I experienced this pressure daily when I led the IRS through one of its most significant modernization initiatives in decades. In 2023, we initiated a comprehensive effort to enhance taxpayer services, improve compliance, and boost operational efficiency. From the outset, accountability was paramount. Every dollar we spent was taxpayer money, meaning each investment had to yield measurable improvements.
Our strategy focused on identifying critical pain points, applying AI practically, measuring the impact, and then building upon those successes. Private companies can adopt this same methodology, provided they understand how to identify and amplify gains from various forms of AI.
The Three Avenues to Value
The most successful organizations will realize returns from AI through three distinct channels:
- General Purpose AI for Daily Productivity
Readily available tools like Large Language Models (LLMs) and agentic workflows can assist employees with initial research and the coordination of simple tasks, thereby freeing up their time. However, the most significant returns in this category will come from training employees to effectively utilize the technology within their specific roles.
Yet, much of tax administration operates with a significantly lower tolerance for error, where even a minor hallucination rate can pose unacceptable risks. The IRS required superior AI tools. As companies increasingly rely on AI for more sensitive processes and entrust it with proprietary information, we will observe a similar critical shift from broad applications to purpose-built systems.
- Domain-Specific Systems for Precision
In areas where factual accuracy is crucial, such as legal research, tax analysis, or medical documentation, domain-specific AI tools will provide companies with a significant competitive edge. These systems are designed using authoritative data sources and incorporate built-in safeguards that substantially reduce hallucinations and enhance reliability. They also offer a quicker return on investment because they are modeled around well-defined workflows and adhere to specific regulatory and security constraints.
Our initial step in the IRS modernization effort using AI was to address the taxpayer service hotline, which suffered from persistent backlogs, extended wait times, and inconsistent responses. We implemented domain-specific AI for response handling, enabling the instant resolution of common inquiries and the efficient routing of complex issues to specialists. Within the first year, average call resolution times decreased from 28 minutes to just three, and millions more calls were answered live.
From utilizing legal contract tools to shorten review periods to employing financial and operational AI to improve planning and supply chain decisions, successful businesses will move beyond retrofitting standard AI models to solve existing problems. Instead, they will apply specialized tools to tackle more intricate challenges.
- Custom AI for Solving Unique Problems
At the IRS, we only invested in custom-built AI when general or domain-specific tools could not meet the required level of data sophistication or compliance standards. For instance, we developed a custom AI solution to analyze patterns across millions of transactions and identify high-risk cases, ultimately preventing and recovering billions in fraud and improper payments during fiscal year 2024.
The common error most organizations make is pursuing custom AI solutions before exploring the first two options. A bespoke AI application not only demands a substantial investment but also presents greater implementation challenges, making it difficult and time-consuming to demonstrate a return on investment.
However, the leaders in the AI race will not necessarily be those with the largest budgets. Instead, they will be the companies that identify the most effective AI use cases through general-purpose and domain-specific applications, generating the ROI and insights necessary to justify custom development.
Iterate and Compound
Achieving consistent returns from AI requires more than just initial successes. Businesses must adopt a dynamic AI strategy, continuously evaluating new possibilities.
As I was departing the IRS, for example, there was growing interest in using AI to replace millions of lines of legacy code written in outdated programming languages. However, the incoming leadership identified a more effective approach: using AI to maintain the existing code. The objective remained the same, but the solution was less risky and more scalable.
Even with ongoing innovation, companies risk falling behind if they do not build upon their AI successes. By implementing a robust AI strategy that encompasses all three types of tools, organizations can consistently develop projects of varying complexity and integrate their capabilities to achieve enterprise-wide advantages.
Today, every dollar spent on AI by organizations is significant. Rather than rushing to integrate AI more rapidly, the companies that will succeed are those that thoughtfully implement purpose-built AI solutions.