Bridging the Gap between AI pilots and profit
The McKinsey 2025 Global Survey reveals that while AI adoption has reached 88%, most enterprises remain trapped in a “scaling gap” where experimentation fails to translate into bottom-line EBIT impact. A small cohort of high performers is pulling ahead by prioritizing agentic AI and fundamental workflow redesign over mere efficiency gains. The future belongs to those who view AI as a catalyst for total organizational transformation rather than a simple plug-and-play productivity tool.
Points clés
- 88% of organizations report regular AI use in 2025, yet two-thirds have failed to scale the technology across the enterprise.
- 62% of companies are experimenting with agentic AI—autonomous systems capable of planning and execution—but only 23% are scaling them.
- Only 39% of respondents report a measurable impact on enterprise-level EBIT despite high adoption rates.
- An elite group of “high performers” (6% of the total) attribute more than 5% of their EBIT to AI initiatives.
- Large companies with revenues over $5B are nearly twice as likely to have fully scaled AI compared to smaller firms.
- High performers are 3.6 times more likely to seek “transformative” change and allocate over 20% of their digital budgets to AI.
- Inaccuracy is the most cited risk, with 33% of organizations reporting negative consequences from AI hallucinations or errors.
- 30% of organizations expect a decrease in workforce size over the next year due to AI integration.
- Software engineers, data engineers, and AI scientists remain the most sought-after talent despite broader workforce anxiety.
- McKinsey surveyed 1,993 participants across 105 nations to compile this data for QuantumBlack.
À retenir
So, everyone is “doing AI,” but almost no one is actually making real money from it yet. It turns out that slapping a chatbot onto a broken 1990s workflow doesn’t magically print cash—who would have thought? If you want to join the elite 6% of high performers, try actually redesigning your business instead of just asking ChatGPT to write your emails. And maybe keep a human in the loop, unless you enjoy explaining to your board why your autonomous agent just hallucinations away your quarterly profits.
Sources
Quiz sur le document: 10 questions
