When AI confronts the reality of processes
As we approach 2025, artificial intelligence is emerging as a recurring theme in executive committees; yet, behind the ambitious rhetoric, operational reality reveals a paradox: AI is being discussed everywhere but still rarely transforms processes. Organizations find themselves at a crossroads where inaction is already tantamount to a choice, with a risk of falling behind that will be difficult to overcome beyond 2027.
This analysis is based on around thirty interviews conducted with experienced decision-makers from the health, industry, services and tech sectors. Their profiles – general managers, data managers, compliance lawyers, innovation managers – share the same tension: reconciling ambition, compliance and value creation in a constrained budgetary context.
The first generation of applications – anomaly detection, predictive maintenance, risk scoring – continues to deliver a solid return on investment as long as it relies on well-established workflows. Where data is clean and governed, these "classic" algorithms run in the background and generate a tangible impact; where datasets remain fragmented or poorly cataloged, AI remains relegated to the status of a decorative proof of concept.
Conversely, generative AI is undergoing massive experimentation, driven by developers and marketing teams, but it struggles to reach the industrialization stage. Budgetary constraints, API costs, regulatory uncertainties, and the lack of a aligned usage framework explain why it remains the preserve of "insiders," while the rest of the organization remains a spectator. An asymmetrical adoption curve is observed: individual effectiveness is tangible, but collective value remains diffuse and rarely measured.
Six key lessons emerge. First, value comes from streamlined, targeted projects that are directly integrated into business tools. Second, the transition from prototype to production proves to be the weakest link: lacking data governance, business support, and change management, many initiatives get lost in a "valley of death" between proof of concept and scaling. Third, sovereignty is addressed pragmatically—segmentation, encryption, and cloud hybridization rather than slogans—but becomes an unavoidable architectural constraint. Fourth, a poorly understood regulatory framework (AI Act, GDPR, NIS2) has a paralyzing effect; some players give up before even starting, fearing the burden of documentation or legal ambiguity. Fifth, the gap between large, tech-savvy corporations and under-resourced SMEs/mid-sized companies is widening; without support, it threatens the economic fabric. Finally, the truly critical expertise is no longer pure data science, but the ability to orchestrate data, risks, processes, and culture.
This orchestration is embodied in six archetypes encountered: the realistic CTO who protects ROI, the strategic compliance lawyer who maps risks, the CEO focused on measurable productivity, the data explorer who drives innovation, the pragmatic CISO who guarantees security, and the operations manager who assesses the utility in the field. Successful projects are those that can align these six voices rather than pitting them against each other.
In contrast, five fundamental trends are shaping the landscape for 2025-2026: the quiet rise of AI agents in back offices, the convergence of analytics and AI for "conversational BI," the emergence of optimized hybrid cloud architectures, the demand for built-in explainability, and operational ethics as a competitive advantage. These movements are perceived less as spectacular revolutions than as a silent industrialization where value lies in the everyday routines.
In light of these observations, five priority areas emerge: industrializing efficiently but to a limited extent, adopting proactive compliance, consolidating a robust data/cloud foundation, developing explainability, and co-creating tools with end users. These areas form a sequential progression: without a reliable database, there is no scalability; without a clear legal framework, there is no legitimacy; without a proven ROI, there is no scaling; without transparency, there is no buy-in; without on-the-ground ownership, there is no sustainable use.
Ultimately, AI will not reward technical sophistication, but organizational discipline: governed data, streamlined industrialization, clear explanations, and proactive business adoption. The next eighteen months represent a critical window; those who structure their initiatives now will gain an invisible but lasting competitive advantage, while inaction risks costing them dearly in an increasingly polarized market.
To learn more: read how unleash the full potential of AI, Why AI culture is becoming a driver of competitiveness and why The adoption of AI depends first and foremost on humans..