The question of the cost of AI is legitimate. It is also, paradoxically, the main obstacle to the transformation of organizations. Wanting to measure the immeasurable is already falling behind.
On the one hand, according to OpenAI’s “State of Enterprise AI 2025” report, ChatGPT Enterprise users save an average of 40 to 60 minutes per active workday. On the other hand, McKinsey assures that the automation of work activities could bring the global economy an annual productivity gain of 0.5 to 3.4% between 2023 and 2040. These figures circulate in all general management presentations. And yet, only 39% of companies report an improvement in their operating income attributable to AI and for most, this impact remains less than 5%.
The gap between promise and measurement is not a technology problem. It’s a problem of method. We apply an accounting tool, in this case ROI, to a transformation which does not obey accounting logic. It’s like trying to measure the value of an information system with an abacus.
The paradox of impossible calculation
To seriously measure the ROI of AI in an organization, we would have to follow each employee task by task, before and after deployment, quarter after quarter and start again, again and again. Because models evolve at an unprecedented speed in the history of professional tools: between two versions of the same model, capabilities can improve by 20 to 30 points on reasoning benchmarks in just a few weeks. The business case lags behind reality.
AI’s biggest gains aren’t in any Excel table. A lawyer who produces a contract analysis in two hours instead of two days does not generate an accounting line. A sales director who prepares a tailor-made pitch in twenty minutes instead of three hours does not trigger any alerts in the reporting tools. A marketing director who cuts her team’s editorial production time in half does not appear in any operational performance indicators. These gains are real, massive, and invisible by construction in traditional measurement systems.
What this actually changes for decision-makers
AI is not a tool with a fixed and universal output. It is a multiplier whose effect depends on who uses it, on what tasks, with what mastery. A CFO who uses it to shorten his monthly reporting from two days to four hours does not have the same ROI as an HR director who prepares his annual interviews in twenty minutes. Looking for a single figure on such a contextual tool is like wanting to measure the value of reading in euros.
This productivity differential is not seen in an ROI calculation. It can be seen in the results, six months later, when the teams that have moved forward have already reorganized their way of working and the others are just starting to build their business case.
The right question to ask yourself
The organizations that are moving the fastest have not waited for a perfect business case but have understood that from a certain point, the question is no longer “is it worth it” but “can we afford not to do it.”
Because while a financial department builds its Excel model to justify the investment, its sales teams lose opportunities in the face of competitors who prepare their pitches with AI. Its lawyers spend two days on analyzes that their counterparts do in two hours. Its managers write their reports by hand while others dictate them from their cars.
Productivity is not what is best measured. These gains are discrete, accumulated, compound. They do not validate any management committee. They win markets. AI is not measured upstream but observed downstream. And those who wait for proof discover the evidence too late.
