The Human Side of the Semantic Layer: Translating Data Meaning Across Teams
Nikos Nikolidakis argues that governance should focus not just on defining business terms, but on keeping their meaning intact as they move between teams. He suggests practical steps like having colleagues apply a definition to the same real decision, and using plain-language explanations before introducing jargon.
- The same word, such as "active project," can mean different things to a budget manager and a delivery team, so a consistent number can still support incompatible decisions.
- Misunderstandings often turn out to be real disagreements over consequences, such as losing funding when a project is declared complete, and need resolving through authority and escalation rules rather than clearer wording.
- Data ownership should be defined by specific decisions, limits on authority, and escalation routes, not vague phrases like "responsible for the data," and translators in teams deserve recognition and time.
Nikos Nikolidakis, who leads the DATAxAI Swarm at GIZ, argues that data governance has to look past the semantic layer's technical job of linking tables and calculations to business concepts. The harder problem is what happens when a definition leaves the model and enters a management meeting or a team's reporting routine. His example is "active projects": a budget manager may count a project as active while payments are still outstanding, while the delivery team considers it finished. Both can nod at the same number and never realize they mean different things. A technically consistent figure can still support incompatible decisions, so the goal, in his framing, is compatible meanings rather than identical data language.
He draws on a few sources to make the point. Beth Bechky's study of engineers, technicians and assemblers on a production floor links misunderstandings to differences in work context, and Paul Carlile's framework separates three situations: those where sharing knowledge is enough, those where interpretations need translating, and those where conflicting interests mean the definition itself has consequences that wording can't fix. He also points to Shulman and colleagues' experiment showing that jargon hampers comprehension even when definitions are supplied, and Cruz and Lombrozo's finding that jargon can make explanations feel more satisfying while making them less understood. Nikolidakis is careful to say the jargon study concerns science communication, not governance programs, but he thinks the risk applies: specialist vocabulary can signal that governance belongs to experts and discourage others from contributing.
His practical proposals are modest and cheap. Ask colleagues from different teams to apply a definition to the same realistic decision, and use the disagreements to sharpen the explanation, the definition, or who has authority to settle it. Lead with plain language such as "knowing where the data came from and what happened to it along the way" before introducing "data lineage", without letting the simplification change the meaning. On ownership, he argues that "responsible for the data" is too vague; an owner should be accountable for named decisions about a defined dataset, with stated limits and a clear escalation route when definitions conflict. He is clear that no tool purchase is needed, though whether organisations will actually run these exercises is left open.