Are 95% of AI Pilots Failing — or Are 75% of Companies Seeing Value? Yes.

By Trailblaze Labs | Published 2025-11-07 | Strategy | 9 min read

If you've felt whiplash from AI headlines, you're not alone. Two major studies tell different stories — but both can be true. Here's how to avoid the failure pile and join the winners.

Welcome to AI's favorite paradox. In one corner: MIT reports that a staggering percentage of AI pilots "fail." In the other: Wharton publishes a study showing that a majority of companies are actually seeing real value from generative AI. Both are credible. Both have data. And both are, somehow, true.

What MIT Actually Found

MIT's research focused on computer vision AI use cases. Their finding that many pilots fail was specifically about cost-effectiveness — where the total cost of implementing AI (hardware, talent, integration) often exceeded the economic value it delivered. The key word: "not yet cost-effective." That's very different from "doesn't work."

What Wharton Actually Found

Wharton's survey covered a broader range of generative AI uses. They found that most companies using gen AI are seeing meaningful returns — but the companies succeeding share specific habits: daily usage that became routine, ROI measured from the start, and leadership that personally owned the initiative.

The Real Lesson

Both studies describe the same elephant from different angles. The pilots that fail usually share common traits: vague scope, no measurement plan, no executive sponsor, and the expectation that AI is a magic wand rather than a tool that needs to be integrated into real workflows.

The ones that succeed picked specific, measurable problems. Usage turned daily — not occasional. ROI was measured from the beginning, not mid-stream. And leadership owned it — CAIO or not, responsibility moved to the C-suite.

Putting it Simply — Is It Time to Get Fit?

Think of AI adoption like fitness: MIT frame says most people buy the gym membership (pilot) and never show up (integration). Wharton frame says people who actually train on a program, log reps, and eat like an adult see results consistently.

If you chase shiny demos, you'll become an anecdote in the "failed" pile. If you build habits, wire to systems, and measure like adults, you'll look a lot like Wharton's winners.