74% of companies say they're getting positive ROI from AI. But only 5% are seeing returns at scale. So what's going wrong with the other 95%?
What's This About?
The Wharton School has been tracking this for three years. Together with GBK Collective, they survey roughly 800 leaders at U.S. companies with $50M+ revenue annually to see whether AI is actually making money.
The 2025 report's headline numbers look pretty impressive. 82% of enterprises use AI weekly, and 74% report positive ROI. Daily usage hit 46%, up 17 percentage points year-over-year. AI isn't an experiment anymore — it's become a daily work tool.
But here's the plot twist. BCG surveyed 1,250 companies worldwide and found that only 5% are achieving AI value at scale. McKinsey research reached similar conclusions — less than 10% have achieved meaningful P&L impact. MIT went even more extreme, reporting that 95% of AI pilots deliver zero ROI.
How do you reconcile "74% positive" with "only 5% succeeding"? Both are true. Most companies are seeing small wins, but scaling those wins to actually move the needle on the bottom line? That's where almost everyone gets stuck.
Why Does This Matter?
The most interesting finding isn't about technology — it's about people. A Wharton study published in HBR in April 2026 revealed that executives and middle managers are living in completely different realities.
45% of executives said their AI ROI was "significantly positive." Middle managers? Only 27% agreed. That's an 18-point gap. The perception of adoption speed was just as stark: 56% of executives said they were moving "much faster" than competitors, while only 28% of middle managers felt the same.
| Perception | Executives (C-Level) | Middle Managers |
|---|---|---|
| AI ROI "significantly positive" | 45% | 27% |
| Adopting "much faster" than competitors | 56% | 28% |
| "Much more positive" about AI | 65% | 39% |
| Describe themselves as "cautious" | 28% | 46% |
Why does this gap exist? Executives use AI for high-level tasks — strategic drafting, decision support. AI works well for these. Middle managers operate in a different world: years-old workflows, teams with uneven technical comfort, output that needs to be consistently right. When AI works, both benefit. When it fails, only one group deals with the fallout.
In short, when AI succeeds, everyone celebrates. When it fails, only middle managers clean up the mess.
Most successful transformations aren't top-down — they're won or lost in the day-to-day realities that managers navigate under real constraints.
— Ana White, Chief People Officer at Lumen
The Playbook: How to Actually Get AI ROI
Synthesizing research from Wharton, BCG, and HBR, five action principles emerge.
- Diagnose before you prescribe
Figure out where your organization actually stands on its AI journey. Don't let executive optimism distort reality. You can't close a gap you can't see. - Co-create the playbook
Don't hand down AI strategy from above. Bring middle managers into roadmap discussions before decisions are made, not after. The people executing need to be part of the design. - Reduce load before adding more
McKinsey found managers spend less than 30% of their time on core leadership tasks. The rest is admin. Asking them to drive AI transformation without reducing that burden is like building a plane while flying it. - Measure readiness, not just adoption
Don't just track "how many people used AI." Make manager confidence and organizational readiness explicit KPIs alongside usage metrics. - Reward honest feedback
Create channels where middle managers can say "this isn't working." Cautious assessments and failed pilots are valuable data, not resistance.
BCG's "Future-Built" Company Traits
The top 5% reinvest AI gains. They allocate 64% of IT budgets to AI, spend 26% more on IT overall, and dedicate 15% of AI budgets to agents. This virtuous cycle produces 5x revenue gains and 3x cost reductions.
