The problem everyone recognises
Here’s a scenario most organisations will recognise. The finance team says revenue last quarter was £4.2m. The sales team says it was £4.6m. Both are pulling from “the system.” Both are confident they’re right. And the board meeting stalls while everyone argues about whose spreadsheet to trust.
The problem isn’t the data. The data is probably all there, sitting across your CRM, ERP, finance system, and half a dozen other platforms. The problem is that nobody has agreed what “revenue” actually means. Does it include VAT?
Does it count at the point of invoice or the point of payment? Does it include that contract that was signed but hasn’t started yet? This is the problem that semantic modelling solves. Not by adding more data, but by adding meaning to the data you already have.
What is a semantic model?
A semantic model defines what your data means — not just how it’s structured. It’s the layer that sits between your raw data and the people (or systems) that need to use it, and it answers the questions that a database schema never will: What is a “customer”? How do we calculate “utilisation”? When we say “active,” active since when?
When you build a semantic model, you’re defining these business concepts once, formally, and making sure they’re applied consistently everywhere — in every dashboard, every report, every query, and every AI model that touches your data.
Think of it as the shared language of your data platform. Without it, every team is translating for themselves, and every translation introduces the risk of getting it wrong.
Image:
Discription of image.
Semantic models vs. Ontologies — what’s the difference?
You’ll sometimes hear “semantic model” and “ontology” used interchangeably. They’re related, but they’re not the same thing.
Semantic modelling is the discipline — it’s the practice of defining what data means and how concepts relate to each other. An ontology is its most rigorous form: a logic-based, machine-readable framework that doesn’t just describe your data but enables systems to reason over it — to infer relationships and facts that were never explicitly stated.
All ontologies are semantic models, but not all semantic models are ontologies. For most organisations, you don’t need to jump straight to a full ontology. A well-built semantic model in Power BI or Microsoft Fabric — one that defines your core measures, relationships, and business logic in a single, governed layer — will transform how your organisation uses data. The important thing is that you build it deliberately, not as an afterthought.
Why it matters — Five reasons
Shared meaning across systems
When you have data flowing in from a CRM, an ERP, a finance system, and a workforce management tool, the same concept — “customer,” “project,” “employee” — is often represented differently in each one. A semantic model maps these to a common definition, so disparate systems can exchange information without ambiguity or misinterpretation. You stop arguing about whose numbers are right because there’s only one version of what those numbers mean.
The connective tissue of data integration
Most traditional data architectures rely on brittle, point-to-point translations — system A feeds system B through a custom pipeline that breaks every time either system changes. A semantic model eliminates this fragility by mapping different data sources to a common conceptual model. Change a source system and you update the mapping, not the entire downstream architecture.
A single source of truth for business concepts
Terms like “revenue,” “utilisation,” “churn,” or “active customer” are defined once, formally, and consistently applied everywhere. This sounds simple but it’s transformative. It ends the fragmentation that makes enterprise data so hard to trust, and it means that when a dashboard says “revenue is £4.2m,” everyone in the room knows exactly what that number includes and excludes.
The foundation for AI that actually works
This is where it gets increasingly important. AI doesn’t work on raw, ambiguous data — or rather, it does, but it produces raw, ambiguous outputs. When you ground your ML pipelines in a semantic layer, models train on data that is consistent, labelled, and contextually understood. The difference between AI that surfaces genuine insight and AI that generates confident-sounding nonsense often comes down to whether the underlying data has been semantically modelled.
Future-proof and extensible
Because meaning is modelled explicitly and independently of any one system, your data platform can absorb new data sources, new business domains, and new use cases without rebuilding from scratch. An acquisition brings three new systems? Map them to the existing semantic model. A new AI use case needs data from five different sources? The semantic layer already knows how they relate. This is how data platforms scale without accumulating technical debt.
What this looks like in practice
We build semantic models for every client engagement, and the approach is always the same: start with the business, not the technology.
That means sitting down with stakeholders and asking the deceptively simple questions. What do you mean by “customer”? How do you define “revenue”? When you say a project is “active,” what triggers that status and what ends it? These conversations are often the first time an organisation has explicitly agreed on its own definitions — and they’re almost always more valuable than anyone expects.
From there, we build the semantic layer into the data platform — typically within Microsoft Fabric or Power BI — so that every dashboard, report, and AI model draws from the same governed set of definitions and measures. When a new data source is added or a new question is asked, the model extends rather than breaks.
In one recent engagement, a client had 10+ core systems with no master data management and reporting built on competing Excel files. There was no single version of the truth and no trust in the numbers. By building a unified semantic model across all sources, we gave every team — from the shop floor to the board — the same trusted view of performance. The conversation shifted from “whose numbers are right?” to “what do these numbers tell us?”
The takeaway
If you’re investing in a data platform — whether that’s Microsoft Fabric, Databricks, or anything else — the semantic model is not a nice-to-have. It’s the thing that determines whether your platform becomes a trusted foundation for decision-making or just another expensive data store that nobody fully trusts.
The good news is that it doesn’t have to be an enormous undertaking. Start with the core concepts that matter most to your business, define them properly, and build from there. The model grows as your platform grows.
The organisations that get this right early don’t just get better dashboards. They get faster decisions, more capable AI, and a data platform that scales with them rather than against them.
