By Greg Woolf, AI RegRisk Think Tank
MIT’s State of AI in Business 2025 report sent shockwaves through markets with its claim that 95% of enterprise AI pilots are failing. Investors seized on the headline. Analysts called it proof that the “AI bubble” is popping. AI-adjacent stocks like Nvidia and Palantir slid in response.
Click Here to Learn More About the AI Readiness
Program for Financial & Wealthtech Firms
But what does “failure” really mean? The study’s definition was narrow: unless a company publicly reported measurable productivity or P&L impact, the pilot was counted as a bust. That’s not the same as saying AI doesn’t work, it’s saying most organizations haven’t crossed the chasm from the experimental Phase 0 into Phase 1 operational efficiencies.
Phase 0 to Phase 2: The Roadmap for AI Adoption
AI doesn’t roll out in one clean leap. It follows phases:
- Phase 0 – Pilots: “Does this even work?” Small experiments, often disconnected from enterprise strategy.
- Phase 1 – Operational Efficiencies: “Can we automate existing tasks and prove ROI?” Early productivity gains show up in workflows, processes, and cost savings.
- Phase 2 – Transformation: “What can AI let us do that we couldn’t before?” Entirely new products, services, and business models emerge.
This framework is critical for understanding both the MIT report and the market’s reaction and why the headlines are misleading.
Gaps in the Study
A closer look at the study methodology raises red flags:
- Thin sample size: 52 interviews, 150 surveys, and a review of press releases. Not exactly a foundation for rewriting the future of markets.
- Skewed lens: Findings leaned heavily on sales and marketing executives, which distorted budget allocations (claiming half of GenAI spend goes to marketing).
- Loose definitions: “Pilot,” “success,” and “implementation” were poorly defined, often collapsing nuanced deployments into black-and-white categories.
As one industry commentator put it, the report was “vibes masquerading as research” which was weaponized by markets looking for reasons to sell.
How Fortune and Forbes Framed It
Two leading business outlets offered deeper analysis:
- Fortune argued that the real problem isn’t technology, but its organizational execution. The lack of leadership buy-in, poor change management, and cultural resistance.
- Forbes went onto say that companies fail because they avoid friction. AI transformation requires re-wiring processes and confronting uncomfortable changes. Skipping that work guarantees disappointment.
In other words: enterprises, not AI, are failing.
So Companies Aren’t Using AI, Right? Wrong
Meanwhile, individuals inside companies aren’t waiting around. MIT’s own data shows that while 40% of firms purchased official AI licenses, 90% of employees are already using AI daily through personal tools. This “shadow AI economy” highlights the paradox: productivity gains are real, but they’re accruing to individuals, not showing up in corporate P&L. It’s no wonder employees grow frustrated with neutered enterprise tools while enjoying cutting-edge consumer models at home. This disconnect is exactly why Phase 0 stalls.
Case Study: Duolingo’s AI-First Breakout
When enterprises get it right, they start repeating the rewards. Duolingo, declared itself an AI-first company is already seeing significant returns. CEO Luis von Ahn underscored the shift: “AI is really transformative for our business. It helps us teach better. It helps us create content a lot faster. It helps us create content that was just infeasible to do before.”
The financial results show how Phase 1 navigation, by using AI to automate course creation and improve operational efficiency, and crossing into Phase 2 transformation, by reinventing the learning experience itself and delivering new products, has achieved unprecedented levels of success:
- Revenue Growth: Q2 2025 revenue surged +41% YoY to $252M, prompting the company to raise its full-year forecast to over $1 billion.
- Profitability: Net income jumped +84% YoY to $45M, while ARPU climbed 6% as users upgraded to the AI-powered Max tier.
- Product Expansion: Duolingo created 148 new courses in one year compared to the 12 years it took to build its first 100, including AI-powered features like video call practice, Explain My Answer, and Roleplay.
The Real Lesson
The headline “95% of AI pilots fail” makes for great market theater, but it misses the truth. Pilots fail because that’s what pilots do. The real challenge isn’t whether AI works, it’s whether organizations have the vision, leadership, and courage to push through friction and redesign themselves around it. A more interesting study would be on Phase 1 operational efficiency projects that are already showing measurable ROI, and Phase 2 transformations where AI enables entirely new products and services.
Conclusion
For AI First companies, the benefits are substantial: revenue growth, profitability gains, and product expansion on a scale that was previously impossible. The real divide isn’t between AI working or failing — it’s between enterprises that figure out adoption and those that don’t.
Greg Woolf is an accomplished innovator and AI strategist with over 20 years of experience in founding and leading AI and data analytics companies. Recognized for his visionary leadership, he has been honored as AI Global IT-CEO of the Year, received the FIMA FinTech Innovation Award, and was a winner of an FDIC Tech Sprint. Currently, he leads the AI Reg-Risk™ Think Tank, advising financial institutions, FinTech companies, and government regulators on leveraging AI within the financial services industry. https://airegrisk.com
