The role of Chief AI Officer is gradually becoming established within organisations. What does the role actually involve, and what role do you play in supporting your clients in this capacity?
This role can only succeed if it is genuinely embedded within the organisation. At Proximus NXT Luxembourg, the position reports directly to the CEO and has the full support of senior management. I am also responsible for internal IT, which gives me an execution team — an essential condition for moving beyond the ideas stage. My role is to coordinate both our internal needs and those of our clients: the use cases we develop internally can then be offered to our clients, as they face many of the same challenges. An AI consulting team also supports them in a field that is evolving at an ever-increasing pace.
General-purpose AI assistants have become extremely powerful, but they do not have access to a company's internal data. The most mature technology today is Retrieval-Augmented Generation, or RAG: internal documents are integrated into an AI system hosted either on-premises or in the cloud, which primarily relies on this internal knowledge when generating responses. The results are much more accurate, and the risk of hallucinations is significantly reduced.
The applications are very practical: drafting a letter, preparing a proposal based on those produced over the past ten years, or, for a municipality, generating specifications based on previously issued tender documents. In all cases, a human reviews and validates the output before it is used.
The first obstacle is data quality. Generative AI does not require structured data and can work very effectively with emails or images. However, this data must be complete, up to date, free of duplicates and aligned across different databases.
Then comes governance: once the data has been cleaned up, it needs to be maintained. When a new circular or regulation is issued, it must be incorporated and the previous version removed.
Finally, an AI project is not simply an IT project. If it is developed without consulting users, there is a risk that it will never be adopted. It needs to be driven jointly by the business and IT teams.
To what extent do digital sovereignty and data protection concerns influence the technology choices of Luxembourg organisations? How can innovation, security and data control be balanced?
Sovereignty is increasingly part of the discussion. Most major AI models are developed in the United States or China, and organisations are understandably cautious about entrusting their data to a cloud environment located outside Europe.
For non-critical data, this may not be an issue. As soon as customer data is involved, however, the GDPR requires the greatest level of caution.
Sovereign solutions do exist. We are partners of Mistral, whose European models can be deployed at the customer's premises or with a trusted provider such as Proximus.
The question to ask is less about absolute performance and more about suitability: is the latest model on the market really necessary for the use case in question? In many cases, a less recent but well-controlled model can meet the requirements perfectly.
Beyond technological considerations, the integration of AI also raises questions around governance and organisational transformation. How should businesses and public administrations adapt the way they operate to support this transformation?
Following the rise of "shadow IT", we are now experiencing "shadow AI": we cannot prevent employees from using these tools, and blocking everything would hinder innovation.
It is therefore necessary to give employees access to the most effective tools while establishing clear guidelines for their use through an AI policy, as we have done ourselves.
Another common mistake is to simply replicate an existing process using AI. The first question should be whether the process still makes sense. Organisations should start with the desired outcome and use AI as an opportunity to rethink the way the process works.
AI adoption depends as much on technology as it does on skills and usage. How do you support organisations in developing their employees' skills and evolving their professional practices?
We are following an approach similar to the one adopted for cloud computing ten years ago, starting with an assessment of each organisation's level of maturity.
We help organisations define their strategy — which model should be used for which application — and then identify relevant use cases. We also organise training workshops to provide everyone with a basic level of AI skills, which is also required under the European Union's Artificial Intelligence Act.
Today's tools enable non-IT users to build their own prototypes, which can then be reviewed and validated by compliance and IT teams before being put into production.
Which developments do you think are most likely to reshape the way AI is used within Luxembourg organisations?
The pace of evolution is simply too fast to make meaningful predictions: the most powerful models change within a matter of months, so organisations need to retain the flexibility to use them whenever the need arises.
Sovereignty, meanwhile, will continue to grow in importance, with more and more models being hosted in Europe.
The most profound change will come from AI agents, which no longer simply provide answers but can make decisions and take action. Humans will need to remain in the loop, and each organisation will have to determine which decisions it is willing to delegate.
My conviction is simple: artificial intelligence is becoming a major issue for businesses, institutions and municipalities, which must gradually gain a better understanding of its uses and opportunities.
