A report by the United Nations Independent International Scientific Panel on Artificial Intelligence raises a question that should directly interest business leaders: who actually captures the value created by AI? Behind this question lies a much broader problem. Companies have invested heavily in accessing technologies, adopting them, and multiplying use cases. But they still have few elements to measure how a gain achieved on a task actually translates into economic performance.

The Preliminary Report of the Independent International Scientific Panel on AI, published on July 1, 2026, constitutes the first independent global scientific assessment produced by this UN Panel. It notably analyzes technological trajectories, economic implications, security, human rights, and governance. Its role is not to prescribe policies, but to provide independent scientific analysis intended to inform decision-making. (United Nations⁠)

On the economic front, the Panel formulates a particularly important distinction. During the presentation of its work in Geneva, Loreto Bravo reminds us that the question is not only about what AI is capable of doing. We must also understand under what conditions it becomes economically useful, who can adopt it, and who captures the value it creates. Researchers also emphasize that the gains already observed on certain tasks are neither automatic nor necessarily converted into overall productivity, better jobs, or widely distributed growth.

This distinction seems fundamental to me for business. It leads to questioning one of the most widely used indicators since the arrival of generative AI: productivity gain.

We may be measuring AI success poorly

For three years, many experiments have been evaluated based on time saved. A task that took an hour now takes only forty minutes. A document that required a day’s work can now be prepared in a few hours. A developer produces certain parts of their code more quickly. A customer service department handles more requests with the same resources.

These gains exist and can be significant. The Panel report does not contest them. However, it reminds us that improving a task alone does not allow us to conclude that economic performance has improved.

To understand why, let’s take the case of a company with 500 employees that deploys an AI system theoretically allowing them to save 20% of the time spent on certain tasks. The figure seems spectacular. It could quickly become an indicator presented to the executive committee to demonstrate the program’s success.

But a much more difficult question immediately appears: where did that 20% go?

Was this time transformed into additional revenue? Production capacity? Cost reduction? Quality improvement? Market acceleration? Innovation? Customer experience enhancement? Available time allowing employees to focus on more complex tasks? Or was it simply absorbed by more meetings, emails, and activity?

The distinction is essential. A company can significantly improve the individual efficiency of thousands of employees without proportionally modifying its collective performance.

The productivity gain of a task therefore constitutes only the beginning of economic reasoning.

Between adoption and value, the missing link is the organization

This observation leads to a second question: how does an individual gain become organizational performance?

I propose representing this transformation through a simple chain:

AI capability → Adoption → Organizational transformation → Performance → Value

The first step depends essentially on technology. Models offer new capabilities: writing, analysis, synthesis, code generation, research, automation, or progressive handling of certain actions.

The second corresponds to their adoption. The company provides tools, develops use cases, trains employees, and measures their usage.

But it’s between adoption and performance that difficulties really begin.

An organization can use AI massively without having transformed its processes. It can add generative assistants to an organization designed before their appearance, while maintaining the same validation circuits, the same indicators, the same responsibilities, and sometimes the same inefficiencies.

In other words, automating an existing organization does not necessarily mean transforming it.

This proposition constitutes my organizational reading of the Panel’s economic conclusions. The report itself emphasizes that the economic benefits of AI depend on adoption conditions and do not automatically translate into aggregate gains.

For business, the consequence is direct. When a six-step process is equipped with artificial intelligence, the first question should perhaps not be how to accelerate each of these steps. It should be to determine whether the six steps are still necessary.

This difference marks the transition from automation to transformation.

AI then becomes less a tool intended to optimize the existing than a means to rethink how work is organized, how decisions are made, and how responsibilities are distributed.

Who actually captures the value created?

The question of value capture is probably one of the most interesting contributions of the Panel’s economic analysis. Loreto Bravo explicitly integrates it among the questions for understanding the economic consequences of artificial intelligence.

For a leader, this question deserves to be applied to each AI program.

Let’s return to the 20% time saved by our 500-employee company. Several scenarios are possible. The company can transform part of this gain into additional capacity and produce more with the same resources. It can improve its margin by reducing certain costs. It can reduce its deadlines and transfer part of the benefit to the customer. It can use the available time to develop new products or strengthen work quality.

But value can also be captured elsewhere.

A portion can be absorbed by the technology provider through licenses, inference costs, or increasing use of its infrastructure. Another can be transferred to customers through price reductions. Yet another can be absorbed by the organization itself if processes do not allow converting time saved into measurable results.

Announced productivity therefore does not tell who actually benefits from the transformation.

This question becomes more important as companies depend on concentrated technological ecosystems. The report precisely identifies market structure, concentration, and distribution among the economic dimensions to be analyzed to understand AI’s effects.

For companies, the question of value creation cannot be separated from that of its distribution. An AI strategy should measure not only what a tool allows saving, but also what proportion of this gain actually remains within the company.

AI’s organizational debt

This difficulty reveals another. As companies accelerate their adoption of artificial intelligence, a gap can appear between their technological capabilities and their organizational capacity to absorb them.

I see this as a form of AI organizational debt.

This concept does not appear in the UN report. It is an analytical framework I propose to analyze what happens when the speed of technological adoption becomes greater than the speed of organizational transformation.

Debt appears gradually. Tools are added without removing old processes. Use cases multiply without a common system for measuring their value. Employees use multiple assistants without overall visibility. Agents progressively obtain access rights to internal applications. Providers multiply. Responsibilities become more difficult to identify. Technology costs increase while their economic contribution sometimes remains difficult to measure. None of these decisions is necessarily problematic when taken in isolation. It is their accumulation that creates a difficulty. The company ends up with a very advanced technological architecture placed on an organizational architecture that has evolved much less.

The parallel with technical debt is useful. In a computer system, rapid solutions accumulated over time can make future transformations more costly and complex. AI organizational debt would function according to a comparable mechanism. The more the organization adds uses without reviewing its processes, responsibilities, and decision-making mechanisms, the more the cost of future transformation increases.

This hypothesis also helps put the current race for adoption into perspective. A company that rapidly deploys dozens of applications is not necessarily more mature than one that focuses on a few strategic processes and precisely measures their economic contribution.

Maturity should therefore no longer be confused with the quantity of AI present in the organization.

We need to change the indicators presented to the executive committee

We then arrive at a very concrete management question. What should we measure?

The number of employees using AI remains useful. The number of use cases as well. Time saved is an important indicator. But none of these elements alone can demonstrate value creation.

It becomes necessary to track the complete chain.

Capability. What new technological possibility have we introduced?

Adoption. Who is actually using it and how frequently?

Transformation. What process, responsibility, or way of working have we modified?

Performance. What operational indicator has improved?

Value. What contribution does this improvement make to the company’s economic or strategic objectives?

This last step is probably the most difficult. It forces us to link AI initiatives to the company’s traditional indicators: revenue, margin, costs, deadlines, customer satisfaction, quality, risk, innovation, or growth capacity.

It also forces us to accept that a widely used project may create little value and that a much more limited use case may instead transform a strategic process.

This new framework then modifies the question asked to the executive committee. It’s no longer just about asking how many employees use artificial intelligence or how many projects have been launched.

We must ask: where is our main breaking point today in the value creation chain?

Is the technology insufficient? Is adoption too low? Have processes not been transformed? Do operational gains exist but are they not captured? Or is the company simply unable to measure them?

The answers lead to very different strategies.

The transition from experimentation to execution

The interest of the UN Scientific Panel’s report lies precisely in its refusal of a deterministic reading of artificial intelligence. Capabilities progress, but their economic consequences depend on the conditions under which they will be adopted, integrated, and distributed. The Panel itself specifies that it produces independent, non-prescriptive scientific assessments, intended to inform decision-making rather than impose answers.

Yet this scientific caution provides companies with an important strategic indication. After several years devoted to experimentation and adoption, the main challenge could now shift to conversion.

How to convert technological capability into organizational transformation? How to convert this transformation into performance? How to convert performance into value? And how to ensure that the company captures a sufficient portion of this value to justify its investments?

It’s at this level that the debate on AI truly becomes a general management issue.

The problem is no longer solely technological. It’s not even just about adoption anymore. It concerns the way the company transforms its processes, decisions, and organization to convert AI’s new capabilities into measurable economic results.

The question to ask leaders could therefore be much simpler than current debates about models and tools:

We have adopted artificial intelligence. But where is the value?

It is probably this question that will distinguish companies that use AI from those that will have truly learned to create value with it.

Artificial intelligence creates capability. Organizations create value.