Artificial intelligence has already entered banks, legal departments, customer relations centers and IT teams. It summarizes contracts, prepares analyses, generates code, assists advisers and examines volumes of data that no human team could process alone.
But the gap between adoption and profitability remains vast.
According to McKinsey, 88% of organizations now use artificial intelligence in at least one function, compared with 78% a year earlier. Yet only 39% report a measurable impact on their operating results at the enterprise level, and that impact most often remains below 5%.
The circle of true champions is much smaller: about 6% of respondents attribute more than 5% of their EBIT to artificial intelligence while claiming to derive meaningful value from it.
Boston Consulting Group comes to nearly the same conclusion: only 5% of companies succeed in creating value at scale through AI, while around 60% observe little or no material benefits yet.
The MIT NANDA 2025 report provides a third point of convergence: after 30 to 40 billion dollars invested in generative AI, only 5% of integrated pilots would produce a measurable impact on the books.
Three studies, three methodologies, but one and the same conclusion: value exists. It remains, however, the exception.
The technology is everywhere, profits much less
The problem does not necessarily lie in the quality of the models.
Most companies began by adding AI tools to their existing ways of working. An employee drafts a report more quickly. A developer produces code in less time. An advisor prepares a response in seconds.
These gains are real, but they do not automatically translate into profits.
One hour saved only yields financial value if it allows handling more cases, selling more, avoiding a costly error or sustainably reducing the cost of an operation.
In many companies, AI thus accelerates isolated tasks without transforming the overall process. Work is faster in one place, but the same controls, validations, manual transfers and bottlenecks remain elsewhere.
That is why the rapid diffusion of licenses has not yet fully translated into productivity or financial results.
The lesson of the “10–20–70”
Boston Consulting Group has been summarizing successful digital transformations for several years with a formula: 10–20–70.
