The term dates from the nineteen-nineties and the tooling has turned over several times, but the shape has not. At the bottom are connections to the sources, a data warehouse if one exists, spreadsheets and operational systems if not. In the middle is the model: tables related by keys, and measures, named calculations such as net revenue or active customers, defined once so that every report shows the same figure. On top are the visuals, the reports and dashboards, and the sharing: who may see which rows, how often the data refreshes, what happens on a phone. Microsoft describes Power BI, the most widely taught tool, as its business analytics platform for connecting, visualising and sharing data, in three parts, a desktop application for building models and reports, a cloud service for publishing and sharing, and mobile apps for reading.
| Layer | What is done there | Where it goes wrong |
|---|---|---|
| Connect and prepare | Load sources, clean types, shape tables | Cleaning done in the report instead of once at the source |
| Model | Relationships, a date table, measures with one definition each | Every report defining revenue its own way |
| Visualise | The chart that fits the question; filters; drill-through | A page of fourteen visuals and no question |
| Share and secure | Workspaces, refresh schedules, row-level security | A manager seeing another region's salaries |
The exams follow the same layers. The PL-300 exam for Power BI analysts weights preparing the data, modelling it and visualising it at roughly a quarter each, with managing and securing the rest, and passes at a scaled 700; the modelling section expects fact and dimension tables, relationships and measures written in its formula language. Google's Business Intelligence certificate covers the same ground on its own tools, with more weight on the pipeline that feeds the model. Both assume SQL and treat data visualisation as one layer of the job rather than the whole of it, which is the adjustment most self-taught dashboard builders need to make.
Division of labour
A small team owns the semantic model, the data quality and the security rules. A wider group builds reports on that model. Reports built outside the model reintroduce the disagreements the model was created to end.
Self-service is the promise every tool makes and the model is what decides whether it is kept. When measures are defined centrally and the relationships are right, a sales manager can build a report by dragging fields and get the correct number; when they are not, self-service multiplies the disagreements that the warehouse was built to end. The practical division of labour is a small team that owns the model and the data quality, and a wide group that builds reports on it, with the visualisation and BI courses serving the second group and the modelling and engineering courses the first.