The $162 Billion Question: Why Big Institutions Keep Buying What Doesn't Work

In the third quarter of 2025, the federal government reported $162 billion in improper payments across 68 programs for fiscal year 2024. That figure, released without ceremony in a routine oversight bulletin, represents roughly 4.6 percent of every dollar the government spends. It is the kind of number that should occasion structural reckoning. Instead, it arrived and departed the public record with the mechanical indifference reserved for weather reports.

What makes this number resonate, however, is what sits alongside it in the institutional record: a Gallup survey, conducted in partnership with the National Bureau of Economic Research, asking senior executives at large companies whether artificial intelligence had improved their organization's labor productivity over the past three years. Eighty-nine percent said no.

Eighty-nine percent.

This is not a marginal finding. It is not a contested reading of fine print. It is a direct question, posed to the people responsible for deploying these systems, and the answer is an almost unanimous verdict that the technology has not moved the needle on what organizations actually measure. And yet capital continues to pour in. Vendors continue to promise transformation. Boards continue to approve部署.

The Procurement Reflex

The AI investment cycle operates on a logic that has almost nothing to do with measured productivity. When a major defense contractor or a federal agency commits to an AI contract, the decision chain runs through procurement officers, compliance reviewers, and senior leadership whose performance metrics include words like "modernization" and "transformation" long before they include anything as mundane as output per worker hour.

This is not unique to government. Corporate procurement follows a remarkably similar architecture. The executive who approves a major AI platform investment is not the one who will be held accountable if the system automates the wrong workflows, generates unusable outputs, or gets quietly abandoned after eighteen months. They will be held accountable for having approved a major AI platform investment.

The survey data on the productivity gap—eighty-nine percent reporting no impact—sits uneasily with the $162 billion improper payments figure not because they measure the same thing, but because they reveal the same institutional pathology. Both are downstream of decision-making structures that separate the people who allocate resources from the people who bear the consequences of how those resources are spent. AI vendors understand this architecture intimately. They sell to the buyer, not the user. They promise transformation, not efficiency.

The Measurement Problem Is Also a Moral Problem

There is a tendency in policy coverage to treat improper payments as an accounting embarrassment—a bookkeeping failure to be corrected by better systems and stricter auditing. That framing is not wrong, but it understates what $162 billion in waste actually represents.

Those are dollars that did not go to the veterans waiting for benefits processing. They are dollars that did not reach the rural clinic that needed equipment upgrades. They are dollars extracted from taxpayers and redistributed to contractors, vendors, and ineligible recipients through systems so loosely governed that the government cannot reliably account for where they went. When an institution cannot distinguish between a legitimate payment and an improper one across one in twenty-five dollars spent, the problem is not technical. It is structural.

And here the AI story intersects: the same institutional architecture that produces $162 billion in improper payments is the architecture deploying AI systems that eight-nine out of every hundred senior leaders say have made no measurable difference to their organizations' productivity. The inefficiency is not incidental. It is the output.

What Reformers Keep Getting Wrong

The standard reform response to government waste is better technology. Better data systems. More sophisticated fraud detection. AI-powered auditing tools. This publication has noted before—and the evidence continues to suggest—that this approach treats the symptom while reinforcing the cause.

The cause is decision-making that is insulated from consequences. When a program administrator faces no meaningful penalty for improper payments, and when an agency CIO faces no meaningful accountability for AI deployments that produce no productivity gains, the rational choice is to keep approving spending that carries the appearance of action without the burden of results.

The Gallup/NBER finding that eighty-nine percent of leaders report no AI productivity impact over three years is not a data point that will appear in the next vendor pitch deck. It will not slow the procurement reflex by a single budget cycle. It will be absorbed by the same institutional machinery that produced it, categorized as a problem for another quarter, and filed alongside the oversight bulletins that nobody reads.

Until that machinery changes, the $162 billion figure will recur. The AI productivity gap will persist. And the gap between institutional investment and institutional output will continue to widen in the way that it always does when accountability is structurally absent.