Data are not universal
“We have plenty of data, but it keeps getting harder to work with it.”
They weren’t talking about systems. They weren’t talking about tools. They were talking about a deeper fatigue — the feeling that everything takes more effort than it should.
That every analysis spawns another debate. That every number comes with a “it depends.” That even when the data are there, no one feels entirely confident using them.
That feeling tends to surface before a company takes a big technological leap. It appears when data stop being silent enablers and instead become constant friction points. And it rarely has anything to do with how much data an organization has; it has to do with how much context they lack.
For years, data have been treated as if they were universal — as if their meaning were self-evident and shared. As if it were enough to capture and store them to start generating value. But data have never worked that way; they always belong to a story: a business process, a decision, a moment in time.
When that story isn’t clear, the data become ambiguous. And when data are ambiguous, people do the only thing they can — they debate them.
Ownership as care, not control
That’s where many organizations begin to grow tired. Not because they don’t use their data enough, but because they work too hard to make sense of them. Each team builds its own interpretation, its own vocabulary, its own mental model, and while no one is entirely wrong, no one is fully aligned either.
For years, DAMA International® has insisted on a principle that looks simple from the outside and uncomfortable from the inside: data are business assets. That means they cannot live only inside systems or technical documentation; they must live in decisions, responsibilities, and conversations.
In the DAMA-DMBOK® framework, data ownership is not just about control, it is about care: knowing who ensures that a data element still makes sense, who can explain what it means, in what context it is used, and why it may not say the same thing everywhere.
Many organizations have discovered that a seemingly simple term like customer can mean different things to different teams, and that this is not a mistake but a reflection of how the business actually operates. The real problem begins when no one has done the work of aligning those meanings, explaining their coexistence, or defining when each version applies.
Without that work, every conversation about data starts from zero. And every new initiative inherits that confusion.
From semantics to lived context
Healthy data governance does not erase differences — it contextualizes them. It makes the implicit explicit, and allows people to understand not only what the data say but why they say it. When that context is shared, work begins to flow more naturally, with far less friction.
This is not just a problem of the present; it is a problem of the future. Organizations that want to scale, automate, or rely more heavily on data-driven decisions cannot depend on tribal knowledge or tacit agreements. The structures of the future — more distributed, faster, more complex — need solid foundations, and those foundations are not technological, they are semantic.
But semantics alone are not enough if they remain theoretical. A shared business context only becomes useful when it is infused with real stories from the organization: actual decisions taken, exceptions resolved, and edge cases that forced teams to think twice. When those cases are captured as definitions, examples, and playbooks, the semantic layer stops being an abstract dictionary and becomes a catalogue of how the business really works. That is exactly what scalable AI needs: not just clean data, but data whose meaning is anchored in real outcomes, so that models and agents can reason with the same context humans use every day.
When the semantic model becomes the backbone
A large home improvement retailer ran into this wall after modernizing its analytics platform in the cloud. Technically, data were centralized and accessible, but every department still calculated “sales”, “inventory”, or “margin” differently, and self-service analytics was constantly blocked by conflicting definitions and slow, inconsistent queries.
The turning point came when they implemented a semantic model on top of their warehouse through a universal semantic layer. Core business metrics were modeled once — with clear names, business definitions, and calculation rules — and then reused everywhere: BI dashboards, spreadsheets, and even natural-language interfaces. The impact was concrete: sub-second queries for most analyses, consistent KPIs across departments, and a foundation ready for AI, where enterprise chatbots and copilots could ask the semantic layer for governed answers instead of improvising SQL on raw tables.
In practice, the semantic model turned governance into something operational: it encoded the organization’s real decisions into a form that both humans and AI systems could trust and reuse.
That is also why DAMA-DMBOK®, newer 3.0 efforts, and frameworks like DCAM increasingly stress real use cases, not just policies on paper. They focus on how decisions are actually made, how exceptions are resolved, and how data are explained when someone new joins the organization. That is the point where governance stops being a document and becomes something alive.
Tools, AI and the semantic gap
In that same spirit, platforms like Atlan do not try to impose a single truth; instead, they help context travel with the data, connecting business definitions to real assets, making ownership visible, and allowing business and technology to collaborate over a common layer. This matters even more for AI, because without a shared semantic foundation, large language models tend to be inconsistent, hallucinate metrics, or leak sensitive signals.
Recent experiments presented at semantic-layer summits and in vendor benchmarks show that LLMs querying raw warehouse schemas can be wrong the majority of the time, while the same models grounded in a governed semantic layer reach near-perfect accuracy and much better determinism. In other words, the same AI becomes “trustworthy” or not depending less on its architecture and more on whether the organization has done its homework on semantics and governance.
When data are well cared for from the beginning, many things stop being uphill battles. Discussions shrink, trust in metrics increases, and new projects no longer need to rebuild from scratch because the shared language is already there. Data start to accompany the organization instead of weighing it down.
It is no coincidence that Gartner predicts that by 2027, 80% of data and analytics governance initiatives will fail, largely because they are not tied to real business outcomes and do not adapt to modern, AI-driven contexts. Many data and AI initiatives do not fail for lack of technology, but because they sit on top of weak governance, incoherent semantics, and a lack of shared context.
Working well with data from the start is not a luxury; it is quiet foresight. It is preparing the ground before building on it. Because data are not universal. And realizing that early changes more than most of us are willing to admit.
To leave you with a few questions
Is your governance program creating trust in data, or just governance theatre — a lot of meetings, policies and committees with no impact on how metrics and models actually behave?