The New York Times recently revisited a problem that has surfaced with nearly every wave of new technology: massive investment without clear returns. In the 1980s, economists called it the productivity paradox. Despite billions spent on personal computers, measurable efficiency gains took years to emerge. Today, the same story is unfolding with generative AI.

According to McKinsey, nearly 80 percent of companies report using generative AI. Yet just as many admit they have seen no significant bottom-line impact. IDC estimates corporate AI spending will nearly double this year to $61.9 billion. At the same time, S&P Global found that 42 percent of AI pilot projects were abandoned in 2024, up sharply from 17 percent the year before. Gartner now predicts AI is sliding toward the “trough of disillusionment,” a stage when early hype fades and organizations grapple with the hard work of proving value.

The Times highlighted case studies that illustrate this paradox. At USAA, after closing down several pilots, the company now uses an AI assistant to support 16,000 customer service workers. Employees like the tool, but leadership acknowledges that the financial payoff is still unclear. Johnson Controls has deployed an app that trims 10 to 15 minutes from a repair call, a modest but useful efficiency gain. And at JPMorgan Chase, roughly 200,000 employees now use an AI assistant tailored to their work. Even so, the bank’s CIO admits they have shut down “probably hundreds” of projects to find what actually works.

These examples are not failures. They are reminders of how difficult it is to quantify the impact of AI in practice. Technical capability does not automatically translate into measurable value.

For government, this lesson is even more urgent. Federal agencies are investing heavily in AI pilots, from call centers at the Department of Veterans Affairs, to predictive maintenance in the Department of Defense, to taxpayer service enhancements at the IRS. Each effort holds promise. But the key question remains: are these projects producing measurable improvements in mission outcomes, or are they simply adding to the list of pilots?

This is not about slowing innovation. It is about asking, in plain terms, is the payoff worth the effort. In other words, is the juice worth the squeeze. To answer that question, we need two types of expertise. We need the builders, the people who design systems and deploy AI. And we need the measurers, the people who know how to track outcomes and determine whether investments are making a difference. Both roles are essential.

The value of measurement is not a new lesson. During World War II, analysts studied bullet holes on aircraft returning from missions. The first instinct was to reinforce the areas with the most damage. But statistician Abraham Wald realized that those planes survived precisely because they were hit in those places. The real vulnerabilities were in the areas with no bullet holes at all, because planes struck there never made it back. That insight changed how aircraft were armored and saved lives. It was a reminder that innovation requires both the builders of the planes and the people who measure how they are used.

Government AI adoption needs the same balance. Builders to push technology forward, and measurers to ensure that each investment produces real and replicable impact. Without both, agencies risk repeating the paradox described in the Times: billions spent without proof of return. With both, AI can move from hype to a proven, measurable asset for mission success.