Article
The real problem with AI isn't the technology. It's that leadership doesn't understand it.
AI: 20% algorithms, 80% leadership
Three of the world's most rigorous research firms measured the same thing in 2025. They reached the same uncomfortable conclusion I reached, and it has nothing to do with technology.
The next competitive gap will not be only between companies that use AI and those that don't.
It will be between leadership teams that understand how AI changes operating models, and those that still treat it as a technology layer.
Most still don't distinguish that difference well. That is the strategic risk.
The data worth facing head-on
When three independent studies, with different methodologies, converge on the same figure, it pays to pay attention.
McKinsey, in its State of AI 2025 (1,993 organizations), found that 88% of companies already use AI in at least one function, but only 39% report any effect on EBIT, and in most cases below 5%.
BCG, in a study of more than 1,250 companies, was blunter: only 5% achieve value from AI at scale, while 60% obtain no material value despite substantial investment.
And MIT, in its GenAI Divide report, found that 95% of generative AI pilots produced no measurable impact on the P&L.
88%
of companies already use AI
McKinsey · 2025
39%
report any effect on EBIT
McKinsey · 2025
5%
capture value at scale
BCG · 2025
95%
of GenAI pilots with no measurable impact
MIT · 2025
Three firms. Three methodologies. One conclusion: adoption is massive, but value is rare.
The right question is not "are we using AI?" Almost everyone is. It is "why do so few capture real value?"
The bottleneck is not the technology
Here is the most important finding, and the most ignored. MIT was explicit: pilots fail because of integration, data and governance gaps, not because of model limitations. As McKinsey summarizes it, AI is 20% algorithms and 80% organizational redesign.
The most revealing data point in the MIT study is indirect: while only 40% of companies have official AI subscriptions, 90% of workers already use personal tools like ChatGPT or Claude for their daily tasks. There is a "shadow AI economy" that works better than corporate systems, and that most leaders neither see nor govern.
Meanwhile, 42% of companies abandoned most of their AI initiatives in 2025, up from only 17% the year before. That is not a slowdown. It is a sign of investing without understanding.
Why the delegation model breaks
For years, leadership operated on the basis of delegation: executives set the direction, technology teams executed, and business units adopted.
With AI, especially GenAI and agentic AI, that model starts to break.
Leaders can still delegate execution. But they can no longer delegate understanding.
What leadership needs to know (and it isn't Python)
This is not about learning to code or tuning neural networks. It is about knowing:
- Where AI creates value and where it creates risk
- What should be built versus bought
- What should be centralized versus federated
- What must be governed
- What should never be automated without human accountability
When leadership cannot ask those questions, trade-offs fail in predictable ways:
- Overinvesting in tools while underinvesting in data
- Launching pilots without a real path to deployment
- Underestimating governance
- Centralizing everything and losing speed, or decentralizing too much and fragmenting
A BCG data point illuminates the last one: AI budgets are overwhelmingly concentrated in sales and marketing, even though ROI is higher in operations and finance. It is exactly the kind of wrong trade-off that happens when decisions are made without fluency.
Where value really lives
What does the 5% that captures value do differently? According to BCG, the winners, the "future-built" companies, stand out for redesigning their workflows and raising their talent's capabilities, not for having better models. The real value of enterprise AI lives in the dependencies most decisions omit: data quality, process redesign, governance, adoption, accountability and operational discipline.
A team can approve an AI strategy. But if it cannot distinguish between a chatbot, a RAG architecture, a predictive model and an autonomous agent, it will struggle to ask the right questions, and it may assume that tools like ChatGPT or enterprise copilots will solve complex business problems on their own.
They won't.
Gartner frames the same blind spot from the top: executive AI literacy does not require coding, but it does demand the strategic intelligence to question assumptions, evaluate risks and align AI with long-term priorities. Without that literacy, initiatives get approved that are not fully understood.
What I have seen in practice
Leading data and AI transformations in environments as different as pharmaceuticals, retail and technology services, the pattern that repeats most is the same. When I implemented predictive models for oncology and vaccines, what separated a successful pilot from one left by the wayside was almost never the algorithm: it was whether the business had redesigned its decision process around the model. The projects that captured value were not the most technically sophisticated, but those where leadership understood that AI changed how the business operated, not just which tool it used.
It is the same conclusion McKinsey, BCG and MIT reached separately. I saw it project by project.
That is the gap that will define the next decade.
Most leadership teams believe they understand AI. The results, often, say otherwise.
Where do you see the biggest gap today: boards, C-level, business units or technology leadership?
Sources
- McKinsey, The State of AI 2025.
- BCG, study on AI value creation across more than 1,250 companies (2025).
- MIT, The GenAI Divide (2025).
- Gartner, on executive AI literacy.
More in this topic