Seamlessly integrating AI at the heart of business processes
From experimentation to industrialization: AI is now emerging as a strategic driver at the core of organizations, bringing both opportunities and new challenges.
For several years, AI has primarily been approached from an experimental perspective. Organizations sought to understand its potential, tested its limits, and developed prototypes to assess opportunities for their activities.
“Last year marked a turning point, ushering in a new phase of technology adoption,” comments Audric Lhoas. “Companies have identified relevant use cases and now want to move to the next step: industrializing these initiatives and integrating them into their operational processes.”
AI is therefore set to become part of organizations’ IT architecture, contributing to improved operational efficiency, decision-making, and customer experience.
Infrastructure as a strategic challenge
However, the growing adoption of AI raises several questions, particularly regarding data confidentiality and resilience. How can organizations ensure they remain in control of their information while avoiding excessive dependence on third-party providers?
“Addressing this question requires considering a fundamental element: the technological infrastructure on which AI solutions rely,” emphasizes Gregory Gruber.
Many solutions today rely on cloud platforms, often provided by American players. “These solutions offer great accessibility and a highly dynamic innovation ecosystem. However, they also raise questions for companies, particularly in terms of technological dependency, cost control, and protection of sensitive data,” he adds.
In this context, technological sovereignty is becoming a major issue, especially in highly regulated sectors. Proximus NXT has notably partnered with Mistral AI to facilitate the large-scale deployment of European AI solutions. “The emergence of European players is an increasingly explored alternative. Models such as those developed by Mistral AI enable more sovereign deployments, particularly in local environments, offering greater control over data and infrastructure,” Gregory Gruber explains.
Building the right environments
However, business transformation does not rely solely on choosing an AI model. “Today, many models are available and continue to evolve rapidly,” explains Audric Lhoas. “The real challenge lies in the ability to integrate these technologies into existing information systems and connect them to company data.”
Organizations must therefore orchestrate multiple use cases and integrate them into their business workflows. This involves deploying dedicated environments, such as Mistral AI Studio, to create, test, and orchestrate AI agents connected to various company data sources.
“The goal is not to rely on a single model capable of doing everything, but to build an architecture that can address a wide range of business needs,” he adds.
When agents work as a team
Rather than relying on a single general-purpose model, companies are moving toward an approach based on multiple specialized agents, each dedicated to a specific task.
“One agent can analyze documents, another can process structured data, and a third can generate recommendations. The output of one becomes the input of another, creating automated processing chains,” describes Gregory Gruber.
The challenge is also to connect these systems to the company’s various data sources—document repositories, CRM systems, data warehouses, or operational systems—to feed and enrich them.
“The use of technologies such as knowledge graphs helps structure this information and enhance AI analysis capabilities,” adds Audric Lhoas.
Identifying the right use cases
However, the success of an AI project depends above all on the right methodology.
“The first step is to define the company’s strategic objectives: improving customer satisfaction, optimizing operational efficiency, or enhancing employee experience,” explains Audric Lhoas. “Based on these objectives, teams can identify potential use cases and select those offering the best balance between business value and technical complexity.”
In practice, many organizations choose to start with a simple and visible use case to quickly demonstrate the value of AI. Some prioritize an “iconic use case”—a cross-functional project capable of creating momentum around AI within the organization.
Measuring the benefits
“Moving from experimentation to production also requires demonstrating the return on investment,” emphasizes Gregory Gruber.
Adopting AI represents a significant investment: infrastructure, technological integration, team support, and process transformation. It is therefore essential to measure the benefits achieved.
These benefits may include improved productivity, faster processing, or higher-quality insights. The challenge is to define indicators that allow results to be assessed over time.
2026: the year of operational AI
After the experimentation phase, artificial intelligence is entering a new stage: its real integration into organizations.
Proximus NXT teams support companies throughout this transformation: establishing robust governance, addressing regulatory requirements, deploying technical solutions, and ensuring long-term follow-up.
“Companies that succeed in this transformation will be those able to combine several dimensions: strong governance, an open technological architecture, a clear data strategy, and a pragmatic approach to use cases,” concludes Gregory Gruber.
