When it comes to enterprise AI, we’re living in a very quantum moment: Two heavyweight institutions take their swings, and instead of landing on different planets, they seem to both be right and wrong at the same time. On one side, you have Massachusetts Institute of Technology (MIT) with a headline-grabbing finding: About 95% of generative AI pilots deliver no measurable return on investment. On the other side, you have The Wharton School at the University of Pennsylvania, reporting that a strong majority of enterprises are already seeing positive ROI and are tracking it formally.
Now, before you chalk this up as “MIT is doom-and-gloom, Wharton is sunshine and rainbows,” let’s pop open the data, talk tone, and get to why in true quantum fashion both studies may be correct — and yet both may be missing key parts of the picture.
(Oh, and for the record: Neither school is an athletic powerhouse — which makes this academic duel easier to watch.)
Two Studies, Two Realities
The MIT Lens
MIT’s ‘The GenAI Divide: State of AI in Business 2025’ study paints a sobering picture: Looking at ~300 publicly disclosed AI programs, 350 senior leader interviews and 150 organizational surveys, their team concluded that only about 5 % of AI pilots achieve meaningful ROI, while 95 % deliver little or nothing.
They highlight root causes: Brittle workflows, inability to scale from pilot to production, and significant mismatch between what is measured (e.g., back-office automation) and where investment is going (often sales/marketing).
In other words: Big AI budgets do not equal big AI returns — at least, not yet.
The Wharton Lens
In contrast, Wharton’s “Growing Up: Navigating Gen AI’s Early Years” (2024) and its follow-up “Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise” (2025) show that usage of Gen AI has surged (72% of firms weekly in 2024 up from 37% in 2023) and ROI tracking is becoming mainstream. Key findings:
- 82% of enterprise leaders use Gen AI at least weekly in 2025.
- Three-quarters of organizations already report positive returns.
- But the study is careful: these are still early days, and the gains are uneven across industry and function.
In short: Yes, many firms are seeing value — and the story isn’t “zero ROI for all.”
How Both Can Be Right
The quantum part kicks in when you realize: MIT and Wharton are looking through different lenses. Their populations, metrics and timelines differ — hence surfacing different truths.
- Different definitions of success/ROI: MIT takes a hard financial lens: “Measurable return via workflows scaled to production.” Wharton takes a broader enterprise-adaptation view: Usage, productivity, pipeline ROI.
- Different maturity levels and sample sets: Wharton’s respondents are often already moving beyond pilot phase; MIT is looking at the broader set, including stalled pilots.
- Different focus areas: MIT emphasizes “why so many fail,” focusing on the 95%. Wharton emphasises “those that are progressing,” and tracks leading-edge firms.
So yes — both can be right: 95% of pilots may indeed deliver little immediate impact (MIT) and a large portion of more advanced firms are seeing meaningful returns and measuring them (Wharton). Parallel realities.
How Both May Be Wrong (or Incomplete)
But the story doesn’t end there — both studies leave gaps, which means both may be partially wrong or at least partially incomplete.
Blind spots in MIT’s study
- If your definition of ROI is only the “big financial return” and you exclude productivity, quality and longer-term gains, you’ll miss value that doesn’t yet hit the P&L.
- Focusing on failed pilots may discount the “fast followers” or agile firms that are succeeding, hence painting a harsher picture than the ecosystem warrants.
Blind spots in Wharton’s study
- Positive ROI reports may rely on self-reported survey data, which can reflect optimism bias.
- “Positive return” may mean incremental productivity, not transformation — so the scale of impact may be modest, not massive.
- The winners may be the early adopters in digital, data-heavy industries; firms in heavier physical industries may lag. Wharton itself shows that across manufacturing/retail, ROI is lower.
In short: Both studies capture important slices of the story — but neither captures the entire film.
Shimmy’s Take
At the end of the day, both MIT and Wharton are capturing moments in time in the early era of AI — particularly “agentic AI” and large-scale enterprise adoption. The path from “pilot” to “production” to “transformation” is still under construction.
Here are the takeaway bullets:
- If your firm is trying to deploy AI and you lean on MIT’s results, you’ll get a strong cautionary tale: Most pilots stall.
- If you lean on Wharton’s results, you’ll see hope: Many firms are already seeing value and moving beyond pilots.
- The reality for your enterprise likely falls between those extremes. You may see early gains in specific functions (e.g., data analytics, document review), but true enterprise-wide transformation remains a “next-wave” outcome.
- Metrics matter. Whether you measure “ROI” via immediate cost savings, process speed, revenue growth, or longer-term capability building will shape your interpretation of success.
- Culture, governance and training — these human factors keep emerging across both studies as the differentiator. Technology isn’t the only bottleneck.
So yes: MIT is right. Wharton is right. But neither has the whole answer yet. That’s the quantum paradox of enterprise AI.
And from where I sit (on the Techstrong decks, with a coffee in hand and the ever-present “beat goes on” echo in my head) — I say: Buckle up. We’re moving out of the pilot epoch and into the scaling epoch. Those firms that treat this not as a surface project but as a platform shift — will win. The rest will look back at these studies with a “we meant to, we tried” shrug.
In the meantime, enjoy the duel. MIT and Wharton may not make conference finals (likely no “March Madness” for either school this year), but in the AI business arena, they’re already tip-off. And for the rest of us? Let’s keep our eyes on court, our hands on the controls, and the scoreboard locked on “progress.”
Shimmy’s Take: Enterprise AI isn’t broken. It’s still early. The ROI isn’t uniform. It will come, but it will come in fits, starts, and re-architectures. Measure what matters, build for scale, and treat this as a marathon — not a one-and-done sprint.
