§ Glossary

Terms heard in the room.

Terms heard in leadership meetings without always sharing a definition. Intentionally short and factual.

Sovereign AI: AI systems whose data, models, and hosting remain under the control of the using organization and its legal framework. An architectural stance, not a brand.
AI Act: European Regulation (EU) 2024/1689 governing the placement on the market and use of AI systems in the EU. Classifies by risk level (prohibited, high, limited, minimal) with differentiated obligations on transparency, human oversight, and conformity. Source ↗
GDPR: General Data Protection Regulation (EU 2016/679). Has governed since 2018 the processing of personal data in the EU: lawfulness, minimization, retention, individual rights, security, breach notification, transfers outside the EU. Source ↗
RAG (Retrieval-Augmented Generation): Architecture that couples an LLM with an external knowledge base. Instead of answering only from its parameters, the model first retrieves relevant passages from the base, then generates a sourced answer. Reduces hallucinations and enables source citation.
Business Intelligence (BI): Set of tools and processes that industrialize the production of regular, trustworthy, shared indicators to steer an organization. Covers collection, consolidation, modeling, and presentation (dashboards, reporting). The foundation of data-driven decision making.
Decision-making foundation: Data infrastructure that consolidates operational sources, rationalizes indicators, and lets leadership and the business share a common view of performance. Includes reference data, dashboards, refresh processes, and governance rules.
CDS Views (SAP): Core Data Services Views - data abstraction layer in SAP HANA and S/4HANA. Allows defining reusable semantic models exposable to reporting (SAC Planning, Analytics Cloud) or applications. Modern SAP standard replacing legacy infosets.
SAC Planning: SAP Analytics Cloud Planning - the planning module of SAP Analytics Cloud. Lets you build budget models, financial scenarios, and operational forecasts with imports from S/4HANA or external files, public dimensions, data actions, and analytical stories.
MLOps: Set of DevOps practices applied to machine learning models: versioning of models and data, continuous deployment, drift monitoring, automated retraining, rollback. Required to move from POC to durable production.
LLM (Large Language Model): Language model trained on a very large textual corpus, able to generate or understand natural language. Examples: GPT (OpenAI), Claude (Anthropic), Mistral, Llama. Performance comes mostly from size (parameters), corpus quality, and post-training alignment.
POC (Proof of Concept): Restricted demonstrator that validates the technical feasibility of a use case without aiming at scale. Useful for resolving a technical doubt - dangerous when it becomes a full project without business framing or value measurement.
Data governance: Organizational framework defining who produces, validates, exposes, and consumes which data. Includes reference data, roles (data owner, data steward), refresh processes, quality controls, and sharing rules. Prerequisite for any reliable BI or AI industrialization.
Executive decision-making: An executive team's ability to understand, measure, and steer the operational and financial performance of an organization. Relies on shared, trustworthy, fresh indicators - otherwise decisions are made blind or relitigated in meetings.
AI hallucination: Generation by an LLM of an answer that sounds plausible but is factually wrong or invented. Causes: parametric knowledge limits, ambiguous prompt, no factual source available. Mitigated by RAG architectures, verification filters, and human review on high-risk uses.
AI copilot: AI system that assists a human in their workflow by suggesting actions or content the human validates. Distinct from an AI agent that executes autonomously. Copilots fit when human judgment remains critical to the deliverable.