This article is the result of a reflection conducted within the AI Club created and led by Eurogroup Consulting.
After the excessive media coverage and the injunction to experiment’AI, a new phase has begun: that of justification. In both COMEX and investment committees, the same question now arises: what return on investment (ROI) can be expected from AI, and on what timeline?
This reading quickly finds its limits. When AI modifies processes, the roles between business and IT, or investment choices, the ROI becomes less an isolated calculation and more a support for arbitration. ROI then becomes a decision-making tool: which investments to pool, which dependencies to accept, which risks to assume, and what level of sovereignty to preserve?
The lands where the king remains tangible
The ROI of AI is not elusive. In several configurations, it is even quite clearly demonstrated. It's even a useful starting point, as these initial results lend credibility to the initiatives, reassure business units, and allow decision-makers to get on board.
The first area is that of uses where The financial gain is directly measurable. logic of quick winFraud detection, reduction of avoidable losses, direct economic impact process optimization. In these cases, the link between model performance and value created can be objectified. The return is visible, fast, and presentable to the investment committee.
The second field is that of already industrialized processes. In sectors like banking, insurance, and manufacturing, AI reinforces existing mechanisms with tangible efficiency gains: reduced processing times, lower costs, automation of various tasks, analyses, and even weak signal detection. Here, the ROI is quickly and decisively observed. It directly supports competitiveness in already pressured environments.
Finally, the third is the one interfaces between support functions and business units. A significant part of the value is not realized in showcase uses, but in reducing costly friction: data reliability, lightening time-consuming tasks, reducing irritants for operational teams, and improving coordination between stakeholders. These gains are less visible, sometimes less easy to monetize immediately, but they streamline the organization and then prepare for more ambitious use cases.
These grounds show that the ROI of AI is neither elusive nor purely theoretical. But they also highlight its limitations: what is measured well at the scale of a use case does not yet say what AI commits to at the organizational level.
From Project ROI to Transformation ROI
The ROI as understood in its classic usage reaches its limits as soon as AI surpasses strictly local gains. The analysis must indeed integrate what AI concretely changes: processes, roles, data, architecture, and scalability. AI does not just allow for productivity gains. It can also improve decision quality, accelerate execution, make operations more reliable, or transform the customer experience. At this stage, reasoning use case by use case is no longer sufficient. It is necessary to understand what AI makes possible at the scale of a process, a function, or a value chain.
The subject becomes that of scaling up and the choices it implies in terms of pooling, prioritizing uses, and investing for the future. An AI use case cannot always be judged in isolation. Its profitability sometimes depends less on immediate gains and more on what it prepares for future uses. It fits into a roadmap that mobilizes common building blocks: data, architecture, security, governance, change management. A case that is not very profitable on its own can be decisive if it accelerates subsequent deployments or contributes to creating the conditions for scaling up.
Thinking Beyond Productivity
A high-performing model is worthless if the organization doesn't know how to capture the value it promises. The ROI of AI cannot be reduced to a simple optimization reflex. It also depends on the company's ability to transform its processes and operating methods to integrate this gain. Without it, the value remains theoretical.
A positive business case guarantees nothing if the organization isn't ready to capture the value. Therefore, it's necessary to examine the concrete conditions for implementation in terms of governance, adoption, business-IT roles, and change management.
This reading makes ROI a common language among decision-makers. It allows for the arbitration of strategic choices What do we finance, why do we finance it, and under what conditions will we capture value? ROI remains useful, but it must be read at the level of the portfolio of uses, shared capabilities, and architectural choices. Its evaluation then depends on the organization's ability to finance, organize, and assume its transformation trajectory.
The ROI of AI: Challenged by Strategic Dependence and the Risk of Falling Behind
As uses are deployed, strategic dependency becomes central: proprietary models, cloud infrastructures, providers, pricing, data investments, architecture, security, governance. These choices accumulate, costs spread, and room for maneuver diminishes. Entering the AI realm means entering an ecosystem where costs, risks, and dependencies must be managed, while ensuring the reversibility of choices and control over the value chain.
In some sectors, the risk is no longer just about a poor ROI. It's a gradual detachment: competitive, technological, or in the value chain. The risk to be assessed is not just financial. A company can lose ground by moving too slowly, or lose its grip by engaging too quickly in a poorly understood ecosystem. The real risk is not a bad ROI, but a loss of position.
Conclusion – Thinking about a transformation pathway
As we’ve understood, the ROI of AI is not limited to a single business case or the sum of profitable experiments. It is seen on several levels: direct gains, scalability, contribution to broader transformation, and the organization's strategic position within its value chain.
From then on, the strategic question arises differently and in these terms: What value trajectory do we want to build, what dependencies are we willing to accept, and what investments must we make to retain control?
Trade-offs are not made blindly; they can be rationalized. To shed light on them, we propose a four-level framework:
Level
Decision question
1 – Direct/Unit Value
Does the use case generate a quantifiable benefit in the short or medium term, perceived by the business and defensible in an investment committee?
Examples: avoided costs, protected revenue, time saved, customer satisfaction, increased quality.
2 – Scaling Up
Does the case prepare the next ones by reducing the cost, timeline, or risk of future deployments? A case that is not profitable on its own can be relevant if it accelerates industrialization.
3 – Strategic Ambition
Does the case contribute to a broader transformation, competitiveness, repositioning, or overall performance? It is no longer judged solely on the basis of a local need, but on its contribution to the company's strategic trajectory.
4 – Strategic Position
Does the case involve structuring choices regarding technological dependence, sovereignty, reversibility, and the risk of falling behind? It is at this level that ROI fully becomes a top management issue.
This framework is not intended to replace the business case, but rather to expand upon it. It allows for each AI use case to be placed within an overall trajectory: immediate gains, scaling up, strategic ambition, and managing dependencies. It is under this condition that ROI becomes a true decision-making tool.
