In 2022, the President and the Secretary of the Treasury asked me to lead the IRS through the “largest technology-enabled transformation” in IRS history. The agency received significant new funding, debuted practical applications of genetic AI, and opened new possibilities for modernization.
As a result, the IRS received numerous proposals from emerging AI solution vendors, but questions remained about how AI fits into our reforms. Because our funding came directly from taxpayer dollars, all AI projects had to make a measurable difference to the people we serve. While much of this research was still in its infancy during my tenure, it revealed lasting lessons about what effective AI implementation requires.
Corporate leaders are currently facing similar implementation challenges. By the end of 2025, 68% of CEOs plan to invest, yet only 15% of executives report improved profitability due to AI even more This year it's AI.
As organizations close out the first quarter, they need to develop a practical AI strategy with integrations that will deliver effective results in the coming quarters. Private companies typically operate in a completely separate world from the public sector, but in some cases they can learn important lessons from government initiatives.
For one, to effectively implement AI, companies can follow the IRS blueprint to identify the sweet spot for risk-based initial fixes to operational issues, rather than abstract transformations.
Establish a “risk window”
Although exciting, the bold all-or-nothing solutions we received from our first AI vendors revealed potentially broader risks that needed to be mitigated. This initially caused some confusion. If AI operates in close proximity to taxpayer data, how can we guarantee the safety of taxpayer data?
We approached this challenge by establishing a workable permission structure to reduce risk. sensing Don't let danger stop us. We have established a tolerance where small solutions can be bundled, tested, refined, and ready to scale to the enterprise level. Starting with a controlled micro-implementation, we helped find one that balances acceptable risk and high reward, uncovering the limits of AI, and showing how it can enhance the role of employees.
Business leaders can establish similar “risk windows” to ensure they don’t default to safe but ineffective AI deployments. Similar to the IRS, executives may believe that solutions that address customer pain points are most promising.
Quickly solve front-line issues
Rather than looking for the simplest application of AI tools, executives can focus on how technology can solve specific persistent problems that plague customers and end users without sacrificing employee know-how.
For the IRS, this meant focusing on issues that were already causing friction, such as the taxpayer hotline. IRS phone lines faced persistent backlogs, long wait times, and inconsistent responses. Instead of waiting for a major communications upgrade, we deployed AI tools to route calls more effectively and reveal accurate responses faster.
Our key KPI for telephone service increased from a historic low of less than 30 percent of calls to a historic high of over 85 percent. Following this win, we implemented similar rapid AI improvements to taxpayer pain points, reducing the backlog of unprocessed paper returns and correspondence from a historic high of 23 million to 24 million in late 2022 to a historic standard of less than 1 million by 2024. We also used backend AI to improve the functionality of our website and app platforms more in two years than in the past 20 years.
Importantly, all of our early AI projects were intentionally intended to move quickly. There was no need to rewrite legacy systems, migrate large data sets, or wait for a multi-year modernization program to complete. However, this rapid pace does not mean replacing human employees. These people were already deeply embedded in the processes we were trying to streamline with AI, so we chose to partner with the technology rather than lose their specific expertise.
To identify the best AI solutions to enhance existing capabilities, leaders must consider the entire scope of customization.
Choose when to customize
Developing a winning AI strategy can no longer rely on a single AI tool. The IRS has recognized that success requires the adoption of three forms of AI.
Off-the-shelf tools have helped executives streamline non-tax administrative tasks, but they lack focused training datasets to help with more difficult investigations. Domain-specific AI solutions helped our phone line improvement program better route niche tax inquiries, but they were no match for the most sensitive decisions.
In these cases, we chose a custom solution. We deployed bespoke case management AI to separate high-risk cases from millions of transactions, ultimately preventing and recovering billions in fraud and improper payments in 2024. However, customized AI solutions can have high initial costs and lack adaptability. This means that company leaders must strive to implement all three levels of customization to be successful across a variety of applications.
Building on early wins
Just because a quick-turn, individual AI use case is initially successful doesn't mean it's the right approach forever.
Rather, solving narrow, functional problems can yield ROI learnings that can inform leaders for the next wave of more complex AI adoption. As these insights compound, executives will develop a deeper knowledge of their organization's risk tolerance and discover new ways in which AI can build on employees' existing knowledge.
AI tools may work automatically, but management cannot put them on autopilot. Instead, you need to work within your company's risk framework to identify a combination of solutions that will immediately make a measurable difference.
