As companies pour billions into artificial intelligence (AI), there is a rising demand from employees and shareholders for concrete results. However, by the end of 2025, just 15% of executives reported that their AI initiatives had boosted profits. To bridge the gap between AI hype and actual returns, we can look to an unexpected source: the Internal Revenue Service (IRS).
While a government agency may not seem like a conventional business, it shares the same urgency to demonstrate progress in AI. I experienced this pressure firsthand while leading the IRS through one of its most ambitious modernization efforts in decades. In 2023, we began deploying AI strategically to enhance taxpayer services, ensure compliance, and improve operational efficiency. From the outset, accountability was paramount. Each expenditure was sourced from taxpayer funds, so every investment had to deliver measurable improvements.
Our strategy centered on pinpointing urgent challenges, applying AI in practical ways, measuring the impact, and building upon successes. Private companies can adopt a similar approach, provided they know how to identify and leverage wins from various AI applications.
The Three Paths to Value
The most successful organizations will realize returns from AI through three distinct pathways:
- General Purpose AI For Daily Productivity
Widely accessible tools like large language models (LLMs) and automated workflows empower employees to conduct preliminary research and manage simple tasks, thus freeing up their time for more complex duties. However, the most significant returns in this area will stem from training employees to optimally implement the technology in their specific roles.
In tax administration, a minimal error margin is critical; even a slight hallucination rate poses unacceptable risks. The IRS required advanced AI tools to minimize these risks. As businesses start trusting AI for sensitive tasks and proprietary information, we will observe a similar shift from generic applications to tailored systems.
- Domain-Specific Systems For Precision
In domains where accuracy is crucial, such as legal research, tax assessments, or medical documentation, specialized AI tools provide a competitive edge. These systems are specifically designed around credible data sources and are equipped with safeguards that significantly reduce errors and enhance reliability. They also yield quicker returns on investment due to their alignment with well-defined workflows and regulatory constraints.
During our initial implementation of AI in the IRS modernization initiative, we tackled issues in the taxpayer service hotline, which suffered from long wait times and inconsistent responses. By employing domain-specific AI for response management, we were able to instantly address common queries and direct more complex issues to specialists. Within the first year, response times plummeted from an average of 28 minutes to just three, allowing millions more calls to be handled live.
From implementing legal contract tools to streamline review times, to leveraging financial AI for enhanced planning and supply chain decisions, successful companies will evolve beyond retrofitting off-the-shelf AI models. Instead, they will deploy specialized solutions to tackle complex challenges effectively.
- Custom AI For Solving Unique Problems
At the IRS, we opted for bespoke AI solutions only when general or domain-specific tools fell short of the necessary data sophistication or compliance standards. We implemented custom case management AI to analyze patterns in millions of transactions and prioritize high-risk cases, ultimately preventing and recovering billions in fraud and improper payments in fiscal year 2024.
A common misstep for organizations is to jump directly into custom AI solutions without first exploring general-purpose and domain-specific options. Such applications not only require significant investment but also present substantial implementation challenges, making ROI difficult and slow to prove.
However, the frontrunners in the AI race won’t necessarily be those with the largest budgets. Instead, they will be organizations that identify the most effective use cases via general-purpose and domain-specific applications, creating the returns and insights needed to justify custom implementations.
Iterate and Compound
Initial victories alone do not ensure sustainable returns from AI. Businesses need to maintain a dynamic AI strategy, consistently re-evaluating their capabilities against new possibilities.
For example, as I was departing from the IRS, there was increasing interest in using AI to overhaul millions of lines of outdated code. However, the new leadership identified a smarter approach: using AI to maintain the existing code. Although the goal remained unchanged, this solution proved less risky and more scalable.
Even with ongoing innovation, companies will lag behind if they fail to build on their AI successes. By fostering a robust AI strategy encompassing all three types of tools, organizations can continuously launch projects of varying complexity and integrate their capabilities for comprehensive benefits.
In today’s landscape, every dollar spent on AI is crucial. Rather than racing to integrate AI hastily, the companies that will thrive are those that thoughtfully implement purpose-built AI solutions.
The views expressed in Fortune.com commentary pieces reflect solely the authors’ opinions and do not necessarily represent the opinions or beliefs of Fortune.
