Our blueprint became to identify urgent pain points, practically apply AI, measure impact, and build from there. Private businesses can follow the same methodology, but only if they understand how to identify and compound wins from different forms of AI.
The most successful organizations will derive returns from AI across three lanes:
Widely available tools like LLMs and agentic workflows can help employees conduct initial research and coordinate simple tasks to free up their personal bandwidth. The biggest returns from this lane, however, will come from training workers to apply the technology in the best way for their roles.
But much of tax administration operates with a vastly lower tolerance for error; one in which even a low hallucination rate can create unacceptable risk. The IRS needed better AI tools. As companies lean on AI for more sensitive processes and trust it with proprietary information, we’ll see them make the same critical shift from broad applications to purpose-built systems.
In areas where factual completeness matters such as legal research, tax analysis, or medical documentation, domain-specific AI tools will give companies a distinct advantage. These systems are engineered around authoritative data sources and contain embedded safeguards that significantly reduce hallucinations and improve reliability. They also deliver faster ROI because they are modeled around well-defined workflows and nuanced regulatory and security constraints.
From employing legal contract tools to cut review time to using financial and operational AI to improve planning and supply chain decisions, successful businesses will no longer retrofit standard AI models to solve persisting problems but instead apply specialized tools to tackle more complex challenges.
The mistake most organizations make is that they jump to this lane before exploring the first two. A custom AI application not only requires a large investment but also poses a greater implementation challenge, which makes proving ROI difficult and slow.
But winners in the AI race won’t always be the companies with the largest budgets. Instead, they will be companies that determine the best AI use cases through general-purpose and domain specific applications which generate the ROI and insights needed to justify custom builds.
Early wins are not enough to establish frequent returns from AI. Instead, businesses must adopt a dynamic AI strategy, continuously pressure-testing against what is newly possible.
As I was leaving the IRS, for example, there was growing momentum to use AI to replace millions of lines of aging code written in outdated languages. But the incoming leadership found a smarter opportunity: use AI to maintain the old code. The objective stayed the same but the solution was less risky and more scalable.
Even with consistent innovation, companies will fall behind if they don’t compound their AI wins. With a strong AI strategy across all three kinds of tools, organizations can consistently develop projects of varying complexity and link their capabilities for enterprise-wide advantages.
Today, every dollar organizations spend on AI counts. Rather than racing to integrate AI faster, the companies that succeed will be those who install purpose-built AI with intention.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
