Companies are spending billions on AI and telling investors it's working. Ask the employees and you get a different answer.
About 40% of workers report receiving what researchers at Stanford and BetterUp are now calling "workslop," AI-generated content that looks polished but lacks substance. Slideshows with no insight. Reports with no analysis. Summaries that miss the point entirely. Work that someone still has to fix, which means it didn't save time. It created more of it.
Each workslop incident takes an average of two hours to resolve. At the salaries workers reported earning, that invisible cleanup cost runs about $186 per month per affected employee. For a 10,000-person organization, that adds up to $9 million a year in productivity losses, not gains.
That number matters because it sits directly alongside the billions companies are spending to generate AI-driven efficiency. The promise was that AI would make organizations leaner and faster. What the data is beginning to show is something different. AI is generating work that creates more work, and nobody is measuring it honestly enough to notice.
The Gap Between What Companies Believe and What the Data Shows
According to KPMG's 2026 Global AI Pulse survey, only 7% of business leaders report having established ROI from their AI investments, even as adoption has become near-universal. That's not a small implementation gap. That's near-universal adoption producing near-universal disappointment, and yet, most organizations are still increasing their AI budgets.
The workers closest to the technology are giving the same report. According to the Upwork Research Institute, 77% say AI has increased their workload. The tool sold as a productivity multiplier is, for most of the people using it, creating more work than it eliminates. Meetings still need to happen. Decisions still need to be made. And now someone also has to check, edit, and fix whatever the AI produced before it goes to the next stage.
The perception problem runs deeper than most leaders realize. A study by METR, an AI research organization, found that experienced developers using AI tools completed tasks 19% slower than those who didn't use them. The same developers believed they were working 20% faster.
That's a nearly 40-percentage-point gap between perceived and actual productivity, in a controlled study, among people whose entire job involves working with technology. If the most technically sophisticated users can't accurately measure what AI is doing to their output, the self-reported productivity gains showing up in company surveys deserve serious scrutiny.
Companies Are Restructuring Around Productivity Gains That Don't Exist Yet
Goldman Sachs projects $1 trillion in AI-related investment globally in 2026, including $581 billion in the United States alone. That spending is accelerating even as companies admit they can't measure what they're getting back.
A KPMG survey helps explain why. It found that 78% of business leaders said demonstrating AI's value to investors and boards was a critical factor driving their strategy, not results. The investment isn't being made because the returns are there. It's being made because leadership needs to be seen making it.
That external pressure has internal consequences. Boards are rewarding AI adoption announcements rather than AI outcomes, which means executives who championed these investments have little incentive to report that they aren't working. The metrics being used to measure AI success are largely self-reported and generated by the same organizations that need them to look good. The result is a feedback loop where spending keeps rising, claims keep coming, and the gap between what's being invested and what's actually returning stays hidden.
That gap is now showing up in the workforce. AI-cited layoffs account for 24% of all announced job cuts so far in 2026, according to Challenger, Gray & Christmas. Companies are eliminating roles while simultaneously admitting they can't measure whether AI justifies those decisions. Many of the workers displaced by these cuts are not being replaced by AI systems doing their jobs. They're being replaced by an assumption that AI will eventually do their jobs.
What the AI Productivity Gap Is Actually Costing Companies
A Harris Poll survey of 900 CEOs conducted for Dataiku found that 80% of CEOs worldwide said their jobs were at risk if their AI strategy failed by year end. 72% said their boards were actively pressuring them to show measurable returns. When job security depends on demonstrating AI progress, reporting that a project isn't working becomes a career risk. So most leaders don't.
That dynamic has a price. Stanford and BetterUp research found that AI-generated output requiring human review and correction costs organizations $9 million per year for every 10,000 employees. Workers are spending significant time fixing AI work that wasn't good enough to use, and that time isn't showing up anywhere in the productivity numbers companies are reporting to their boards.
The problem runs deeper when you look at what happens to AI projects that aren't delivering. Emergn surveyed 700 senior business leaders and found that only 30% considered ending an underperforming AI initiative a normal and acceptable decision. Nearly a quarter of senior leaders admitted they were reluctant to acknowledge when an AI project had failed. As a result, those projects keep running, costing U.S. companies an average of 2.4% of annual revenue.
The workforce decisions built on top of those failing projects are proving more expensive. Forrester's 2026 Future of Work research found that 55% of employers regret layoffs made for AI-related reasons. More than 30% of U.S. hiring managers who eliminated roles after adopting AI have since reinstated those positions and even spent more than they saved through the original cuts.
The problem isn't that companies are investing in AI. It's that most are making workforce and budget decisions before they have honest data on what the technology is actually doing. That's the gap worth closing first.
Companies getting real returns from AI aren't deploying it everywhere and hoping for the best. They're identifying the specific tasks where AI demonstrably reduces time or improves output, starting there, measuring carefully, and scaling only what the evidence supports. Broad deployment without a clear use case produces exactly what the data shows: high adoption numbers and negligible returns. The organizations seeing results are treating each implementation as a test, not an announcement.
77% of workers say AI has increased their workload according to the Upwork Research Institute. A significant reason is that most organizations deploy AI tools without giving workers the structured training to use them effectively. Workers who understand how to apply AI to their specific tasks report meaningfully better outcomes than those handed a tool and expected to figure it out. Upskilling existing workers before deployment is also more cost-effective than replacing them with an assumption that AI will fill the gap.
Most organizations have no way of knowing whether AI is working because they never measured productivity before they deployed it. Without a baseline, any claim about AI-driven efficiency gains is essentially unverifiable, which is partly why self-reported productivity numbers are so disconnected from controlled study results. Organizations should establish clear performance baselines at the task and role level before rolling out AI tools, then measure actual output against those baselines at defined intervals. That's the only way to know whether the investment is returning anything real.
Adoption rates, tool usage statistics, and the number of AI prompts submitted per day are not productivity metrics. They measure activity. What matters is whether work is getting done faster, with fewer errors, and at lower cost than before. Leaders should distinguish clearly between AI activity and AI outcomes in how they report to boards, and boards should stop rewarding adoption announcements as though they were results. The pressure to show AI progress is driving the gap between what companies claim and what their workers actually experience. Changing what gets measured at the top changes what gets prioritized below it.
This article was originally published on Forbes.com