The interoperability of technological systems implies that they speak the same language. This is not the case. Another example of a key layer where Luxembourg could play a pioneering role, even if bricks exist left and right. (Photo: Shutterstock)

The interoperability of technological systems implies that they speak the same language. This is not the case. Another example of a key layer where Luxembourg could play a pioneering role, even if bricks exist left and right. (Photo: Shutterstock)

This is a strategic moment. Europe is standardising, regulation is intensifying, AI is pushing to make the implicit explicit. Luxembourg has built its position on cross-border legal and operational orchestration. The next step could be semantic orchestration. In digital finance, whoever structures the language structures part of the market. Long live ontology, adopted by the biggest players.

Luxembourg talks a lot about AI, regtech and the competitiveness of its financial centre. But what is still missing from the public debate, and often from transformation programmes, is a question that is more fundamental than the choice of a model or cloud provider: does the country have a common, formalised language to describe its finance? Without shared definitions - of what is an exposure, a fund, a holder, a risk, a market event or ESG data - automation promises a great deal and sometimes delivers operational fragility. A Luxembourg financial ontology would be a concrete, collective and exportable first step towards putting Luxembourg back on the European, or even global, map.

An ontology, in the sense of the information sciences, is not a philosophical speculation. It is a formal representation of knowledge about a domain: a set of concepts, their properties, the relationships that link them and, increasingly, the actions that transform them. Where a database aligns tables, an ontology makes meaning explicit: it defines what the company or a sector means by “client”, “beneficial owner”, “assets under management”, “operational incident”, “sustainable fund” or “indirect exposure”. It links these objects, encodes consistency constraints and, in its most advanced form, even enables automatic inferences to be made. It is precisely this semantic layer that then enables analytical tools and AI to be plugged in without each project reinterpreting the same words.

It is essential here to distinguish between ontology, data model and taxonomy. The European Banking Authority’s Data Point Model (DPM) [now in 2.0] or the ISO 20022 standard are powerful structuring frameworks, but they remain mainly models or dictionaries for organising data and messages. An ontology, on the other hand, provides a transversal link between concepts, explains their logical dependencies and enables navigation between domains. The nuance is technical, but decisive: it conditions the ability to automate without ambiguity.

What’s changing today is that ontology is no longer just a fancy for architects. The technology giants have industrialised it to run their platforms. Google has popularised the “entities and relationships” logic through its Knowledge Graph, the foundation of a search that includes concepts rather than words. This approach does more than just classify: it disambiguates, connects, infers and maintains a coherent repository on a very large scale.

Amazon, for its part, has had to organise the infinity of commerce: products whose attributes and relationships vary endlessly. The challenge is as much semantic as volumetric. Without stable definitions of concepts such as “compatible”, “certified”, “refillable” or “variant”, search, recommendation and compliance become fragile. Knowledge graphs and automated fact extraction systems aim precisely to industrialise this structuring of meaning.

At Netflix, the subject came home to roost: when the same concepts - work, season, talent, broadcast rights - are modelled differently in different systems, integration becomes costly and automation uncertain. Adopting a common metamodel meets an imperative for consistency and scalability.

In industry, Siemens illustrates another dimension: that of the digital twin. A supply chain or an industrial park cannot be simulated without a formal model of objects, relationships and events. In this case, the ontology determines the ability to reason about reality. Industrial ontologies are often based on high-level standards to avoid local silos.

Finally, Palantir Technologies has put forward an approach where the ontology links data, business rules and operational actions. Objects (contracts, stocks, entities) are formally defined, as are the possible actions on them. The aim is not just to analyse, but to act within a traceable framework. This development sheds light on a crucial point: generative AI can assist without a formal ontology; on the other hand, regulated or audited decision-making AI requires legally stable concepts.

This plea does not aim to invent a parallel Luxembourg. It is based on what already exists in regulated finance. The EBA’s Data Point Model structures the concepts of prudential reporting. The European Central Bank’s BIRD programme harmonises definitions and requirements. The AnaCredit project imposes a standardised granularity for credit data. ISO 20022 is based on a conceptual business model for financial services. The Financial Industry Business Ontology (FIBO) even proposes a formal, ready-to-use ontology.

These initiatives show that the battle for digital finance is not being fought in the algorithms alone, but in the semantic layers. They also show that Europe is making progress, but in a segmented way: prudential reporting on the one hand, payments on the other, financial markets elsewhere.

A Luxembourg financial ontology should neither compete with these standards nor be isolated from them. “Sovereign” does not mean autarkic. It would mean a capacity for orchestration: aligning, integrating and making coherent the existing frameworks in the Luxembourg ecosystem. Sovereignty here would be a capacity for coordination and export, not fragmentation.

There remains the question of cost and governance. Yes, such a project will require arbitration between players, ongoing maintenance and a real political effort. But these costs already exist, scattered across integration projects, diverging ESG interpretations and the ongoing redefinition of internal guidelines. The alternative is not cost versus no cost; it is between repeated fragmentation and mutualised structuring.

Concretely, a pragmatic approach would be to start with a minimal core: entity, control relationship, instrument, exposure, event. Align it explicitly with DPM, BIRD, ISO 20022 and FIBO. Test on a pilot case - for example ESG reporting or cross-border fund distribution - before any extension. No grand semantic evening, but versioned, open and evolutionary experimentation. All this in a combo between the CSSF and the CNPD for reasons that are intuitively understandable.

The usefulness, in Luxembourg, is threefold. Reduce regulatory and integration frictions. Accelerate the development of exportable regtech. Secure the critical uses of AI. In an increasingly automated world of finance, the ability to define concepts precisely is becoming a competitive advantage.

The main risk is well known: producing a theoretical ontology that is not adopted or is captured by a few major players. The answer lies in inclusive, modular and transparent governance.