AI is no longer a promise. Most organizations have now moved beyond discovery and early curiosity. As their AI maturity grows, a new reality is emerging: the challenge is no longer simply to launch initiatives. It is to organize the company and create the appropriate system so AI can actually work in day-to-day operations.
After a first discussion focused on why AI is transforming organizations, and a second one centered on why execution remains difficult, this third roundtable marked another shift. The conversation moved from experimentation and scale to a more structural question: what kind of organization is actually needed to make AI work?
The state of play
On April 2nd, Intys hosted a third cross-sector AI executive roundtable, gathering leaders from various sectors to discuss this new challenge.
The question is no longer only: Can we experiment? or How do we scale it ? It is : How do we operate AI in a coherent, sustainable and well-governed way?
Across the exchanges, one point became particularly clear: AI should not be treated as a separate topic. It has to be integrated into the broader operating model of the organization and connected to real business priorities, constraints and processes.
That changes everything.
To move from isolated initiatives to sustainable value, organizations need a real AI Operating Model. Not a perfect one, because no universal model exists, but a coherent one built around deliberate and context-related choices.
The discussion highlighted five key dimensions that structure this model:
- organization,
- governance,
- roles and capabilities,
- processes,
- platform.
Together, these dimensions define whether AI remains fragmented or becomes a real transformation lever.
There is no perfect model, only explicit choices.
No single winning model exists. Some organizations are building around centralized approach, with a lab or a center of excellence. Others are choosing more hybrid models, where business, IT and transformation functions work together. Others still are trying a bottom-up approach, while keeping control over sensitive use cases.
What matters is not to copy a model. it is to make explicit choices.
As soon as AI use cases multiply, ambiguity appears. If nobody clearly knows who leads, who supports, and who decides, the project slows down. An inefficiency appears not because they lack ideas, but because ownership and execution logics are too vague.
The real debate is therefore not centralization versus decentralization in abstract terms. It is how to find the right balance between coherence and autonomy, speed and control, bottom-up and top-down approaches.
“There are a number of decisions to be made, explicit choices, regarding organization, governance, roles and responsibilities, processes, and the platform. When it comes to turning AI ambition into an organizational reality, you have to address these elements.”
- Loïc Decaux, Head of AI Lab @Intys
Governance, risk and accountability are not side topics.
The roundtable also confirmed that once AI moves closer to operational processes, governance becomes unavoidable. Organizations have to decide how they prioritize initiatives, how they arbitrate between value and risk, and how they create enough structure without killing creation.
Participants pointed to a recurring tension: too much control blocks, too much freedom blurs. That tension is now central to AI execution. A rigid governance model can slow experimentation. But an overly free approach creates shadow initiatives, duplication, inconsistent practices and growing exposure on data, compliance and security topics.
“We need to make sure that people see the designated playground. That they can interact and connect with one another (...).”
- Laurent Bourgois, Head of Business Solutions @Solidaris Brabant
The challenge is therefore not to choose between control and experimentation. It is to design a framework that allows both. That includes clear prioritization mechanisms, explicit risk ownership, defined decision rights, and practical spaces where staff can test and learn within adapted sensitive boundaries.
Shadow IT brings out a deeper issue which is whether organizations are capable of creating a safe and clear environment for experimentation. In practice, many are still trying to find the right balance between training, access control and responsible usage.
Human validation also remains essential. AI is not a tool that replaces accountability, but as a support compatible with security, regulatory constraints and human responsibility.
Culture, capabilities and adoption are becoming the real differentiators.
In the second roundtable, the human dimension was central. The third roundtable discussion confirmed that the current barriers are organizational and cultural. Leaders need to create direction without overcontrolling, and employees need to understand what AI brings, where the risks are, and how to use it responsibly.
This is why training and culture adaptation came back during the roundtable. What was described was not generic awareness-building, but very practical capability development: broad training for general users and more advanced support for developers and project managers. AI adoption will not happen through tool access alone. It requires confidence, understanding and repeated learning loops.
“We train our project managers (...) we equip them with the tools they need (...) We tell them “Go back to your departments and see what can be done”.”
- Arthur Le Paige, AI Project Manager @Solidaris Brabant
The discussion also highlighted an important cultural message: AI should not be framed as a substitute, but as an assistant that augments human judgment. The more organizations demystify the tools, clarify the rules of use, and make experimentation feel safe, the more they reduce resistance and create the conditions for meaningful adoption.
“We want to make people become active agents of change (...) so that, when we introduce an AI agent, it does not scare them, and they are able to identify areas for improvement and provide feedback.”
- Guillaume Meeus, Head of Transformation @Athora
From use case to process transformation.
Another lesson from the roundtable is that AI creates value only when it is embedded into real work. This remains one of the biggest execution gaps in practice.
It is easy to imagine use cases. It is much harder to redesign a process so that AI actually improves the way work gets done. It requires rethinking tasks, clarifying points, assigning human oversight, and sometimes challenging the process itself from scratch.
“New technologies were adopted by those who had the vision and knew what they wanted to do with them.”
- Alain Ejzyn, prof. Digital Strategy @Ichec Brussels School of Management
This is why the question of who is truly responsible for transforming a process with AI is so important. Without that bridge between business reality, process understanding and technological possibilities, many AI efforts fail. This is also why roles such as AI Business Analyst, Business Process Owner, AI Lead, and AI champions are becoming increasingly important. They connect ambition to execution and turn initiatives into transformation.
Platform choice matters, but not at the heart of the issue.
The discussion also addressed the platforms organizations are using. But here again, the deeper question was how these tools are governed, how access is structured, how they connect to the existing systems, and where to draw the line between automation and AI.
That distinction matters. Automation is usually better suited to rigid, repeatable and highly reliable processes. AI brings more flexibility and adaptability, but also more uncertainty. Organizations therefore need to make more conscious trade-offs depending on the nature of the task, the expected level of reliability, and the degree of acceptable risk.
Mindset is becoming the real operating lever.
If one theme stood above the others, it was mindset. Participants pointed to a recurring challenge: organizations still approach AI with pressure for immediate returns, and limited tolerance for failure. But AI rarely works well in that environment.
“The unique characteristics of AI projects require a shift in mindset regarding risk-taking and the time allocated.”
- Arthur Le Paige, AI Project Manager @Solidaris Brabant
It needs short learning cycles, iterative process, room for testing, and a more agile relationship to uncertainty. Not every initiative will succeed. Not every use case will scale. And that is precisely why organizations need a model that makes it possible to test, learn, prioritize and adapt without shutting down initiatives. In that sense, the real differentiator is increasingly cultural.
“We've been working in a waterfall model, and now we're switching to an agile approach.”
- Guillaume Meeus, Head of Transformation @Athora
What’s now and what’s after ?
If the second roundtable was about moving from POCs to scale, this third roundtable addresses the organizational conditions to make AI repeatable, governable and valuable.
This means clarifying who owns the transformation by :
- building governance that controls without blocking.
- investing in roles, capabilities and training.
- redesigning processes, not only using tools.
- treating AI as part of the company’s operating model, not as a separate track.
This is precisely where our approach at Intys is built around.
- Inspire: align ambition with business priorities.
- Innovate: challenge processes, not just tools.
- Integrate: embed AI into governance, operations and day-to-day work.
- Impact: turn experimentation into measurable business added-value.
The question companies now face is no longer whether AI matters, but whether their organizational structure allows them to make it work.
If you would like to explore the subject deeper, or confront your organizational reality, we would be pleased to exchange with you !
Loïc Decaux - Head of Research, Development & Innovation Lead of AI Lab
Luis Parisot - Partner Intys