Enterprise AI at the Decision Point: A U.S.-Focused Systematic Review of Productivity, ROI, Workforce Change, and Governance
This systematic review of U.S. enterprise AI adoption from 2022 to 2026 challenges the prevailing narratives of guaranteed financial returns and widespread workforce substitution, revealing instead that productivity gains are highly task-dependent, independently verified ROI is absent, and critical evidence gaps currently prevent organizations from making fully informed, governance-aligned decisions regarding large-scale deployment or workforce restructuring.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the last few years, a new kind of computer program has moved from science fiction into the daily work of offices and factories across the United States. These are not the simple chatbots of the past that only answer questions; they are intelligent assistants capable of drafting emails, writing code, and even taking actions within complex business systems. Companies are investing billions to install them, driven by two powerful promises. The first is that these tools will make workers significantly more productive, generating substantial financial returns. The second is that they will allow organizations to do the same amount of work with fewer human employees, effectively replacing human labor. These expectations are shaping major decisions about budgets, hiring, and the future structure of the workforce. But while the technology is advancing rapidly, the proof that it delivers on these specific promises has remained unclear.
A new systematic review by researchers at the University of Houston, led by Audrey Rah and David W. Johnson, cuts through the marketing and the speculation to examine what is actually happening on the ground. The team did not rely on vendor claims or theoretical models. Instead, they gathered and analyzed 61 high-quality studies, government reports, and independent experiments published between January 2022 and September 2026. Their goal was to determine if the evidence supports the idea that these tools are ready for mass adoption, or if the reality is more complicated. They looked for hard numbers on how much money companies are actually spending, how much time is truly saved, whether financial profits have materialized, and if the tools are causing the workforce reductions that many leaders anticipate.
The findings challenge the most optimistic narratives. While the technology is undeniably useful in specific situations, the review found that the benefits are uneven and highly dependent on the task at hand. In customer service, for instance, workers using these tools resolved about 15 percent more issues per hour, and software developers completed roughly 26 percent more coding tasks. However, these gains did not happen everywhere. In other settings, the tools produced no improvement at all, or even caused performance to drop. The researchers found that the technology works best when the work is routine and well-defined, but it struggles with complex, nuanced problems that require deep human judgment.
Perhaps more importantly for business leaders, the review found no independent evidence that these tools have generated a measurable financial return on investment for named enterprise products. While some companies report high profits, these figures often come from models or self-reported surveys rather than audited financial records. The gap between the price of a software license and the total cost of running it—including security, training, and the time needed to fix errors—remains largely unmeasured. The researchers could not find a single case where a company could trace a specific dollar amount spent on AI directly to a verified reduction in staff and a confirmed increase in profit.
The question of workforce reduction is where the evidence is most fragile. The review analyzed data from the U.S. Census Bureau and found that while 18 percent of American firms are using AI, the vast majority—66 percent—are using it only to help workers do their jobs better, not to replace them. Only 2 percent of firms reported any AI-related decrease in employment. Even among large organizations that have adopted advanced "agentic" systems capable of taking actions, reports of workforce cuts did not correlate with higher financial returns. In fact, in many cases, companies that reduced staff did not see better profits than those that kept their teams intact. The data suggests that when companies do cut jobs, it is often due to broader economic pressures or strategic shifts, rather than a direct result of the AI doing the work of a human.
The study also highlights a hidden risk: the gap between what a tool can produce and what a human can verify. These systems can generate professional-looking work quickly, but if the human reviewing the work lacks the expertise to spot errors, mistakes can slip through unnoticed. This creates a situation where productivity appears to rise while quality quietly declines. The researchers noted that for these tools to be safe and effective, human oversight must remain strong, with clear rules about who is responsible for the final output. Without this, the promise of automation can turn into a liability.
Ultimately, the review concludes that the evidence is not yet sufficient to justify sweeping, irreversible decisions like massive layoffs or enterprise-wide budget expansions based solely on the hope of AI-driven efficiency. The technology is real, and it is changing how work gets done, but it is not a magic switch that guarantees profit or replaces human capability. The path forward, the authors suggest, requires a more cautious approach. Organizations should treat these tools as powerful assistants that need careful management, rigorous testing, and clear metrics before they are scaled up. Decisions about the future of the workforce should be based on verified results, not on projections or the fear of being left behind. Until the data shows a clear, proven link between AI spending and realized financial value, the most prudent strategy is to move forward with measured steps, ensuring that the technology serves the organization rather than the other way around.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.