Here's a number that should stop every executive in their tracks: 95% of enterprise generative AI projects show no measurable financial return within six months. Not "underwhelming returns." Not "needs more time." No measurable return at all.
This isn't speculation from an AI skeptic's blog post. It comes from MIT's NANDA research group, and it's been echoed by Harvard Business Review, Deloitte's State of AI Enterprise report, and a growing pile of industry data that paints a picture most vendors would rather you didn't see.
The Gap Nobody Planned For
IDC projects a $5.5 trillion loss from AI skills shortages by 2026. Read that again. Trillion, with a T. And yet, when you look at how most companies approach AI adoption, it's tool-first. Buy the license, roll it out, hope for the best. When the results don't materialize, blame the technology.
The problem isn't the technology. The problem is that organizations are trying to bolt AI onto teams that don't understand what AI actually does, what it can't do, and how to tell the difference. You wouldn't hand someone the keys to an industrial robot without training. But that's effectively what's happening with AI across most enterprises.
The U.S. Department of Labor recognized this in February 2026, releasing a formal AI Literacy Framework built around five core competencies: understanding AI principles, exploring AI uses, directing AI effectively, evaluating outputs, and using AI responsibly. It's voluntary, but the signal is clear. Even the federal government sees that the skills gap is the bottleneck, not the models.
The "AI Washing" Problem
Sam Altman, yes, the CEO of OpenAI, admitted in February 2026 that "there's some AI washing where people are blaming AI for layoffs they would otherwise do." When the guy selling the shovels tells you some people are just digging for show, you should pay attention.
Here's the pattern: a company announces a big AI initiative. Stock goes up. Then comes the "restructuring," layoffs branded as "AI-driven transformation." Block cut 40% of its workforce this week, and the stock jumped 24%. The markets rewarded the narrative, not the results.
Meanwhile, an NBER study found that 90% of C-suite executives reported no actual employment impact from AI. The gap between the story companies tell investors and what's actually happening on the ground is enormous.
What Actually Works
IBM is doing something interesting. Instead of cutting junior roles, they're tripling entry-level hiring in 2026, but redesigning those roles entirely. Less repetitive coding, more client engagement, more product development, more of the work that requires judgment, empathy, and context that AI genuinely can't do well. IBM isn't betting on replacing people with AI. They're betting on people who understand AI.
This is the counter-narrative that doesn't make headlines because it doesn't trigger fear. "Company invests in training people" doesn't get the same clicks as "AI will take your job in 18 months." But it's the approach that actually generates ROI.
The companies getting value from AI share a pattern: they invest in understanding before implementation. They don't start with "what tool should we buy?" They start with "what problem are we solving, and does AI actually solve it better than the alternatives?"
The Real Question
Forrester predicts enterprises will defer 25% of planned 2026 AI spending into 2027. That's not a failure of AI. It's a delayed recognition that throwing money at tools without building understanding is expensive.
The 5% of projects that do work aren't using fundamentally different technology. They're being run by teams that understand what they're working with. Teams that can distinguish between what AI does well (pattern recognition, text generation, data synthesis) and what it doesn't (reasoning, judgment, context-dependent decision-making).
If you're planning an AI initiative, don't start with the vendor pitch deck. Start with your team's understanding. Can they evaluate an AI output? Can they spot a hallucination? Do they know when to trust the model and when to override it? If the answer is no, your project is already in the 95%.