2.1.1. Identifying Data Sources and Connecting to Shared Semantic Models
💡 First Principle: The best source is the most trusted, most upstream copy of the data that someone else already maintains. Before building a new import from raw tables, check whether a curated semantic model or a Fabric lakehouse already holds what you need; reusing it gives you consistent numbers and less work.
Power BI Desktop's Get data offers hundreds of connectors, which fall into a few families you should recognize by scenario:
- Files: Excel, CSV/text, XML, JSON, PDF, Parquet. The Folder and SharePoint folder connectors combine many files with the same structure (monthly exports) into one table using a sample-file transformation.
- Databases: SQL Server, Azure SQL Database, Oracle, PostgreSQL, and others. These usually support query folding and DirectQuery.
- Microsoft Fabric and Power Platform: lakehouses, warehouses, and semantic models from the OneLake catalog, plus Dataverse for Dynamics 365 and Power Apps data, and dataflows.
- Online services and web: SharePoint Online lists, Salesforce, Google Analytics, web pages, OData feeds, and REST APIs.
Connecting to a shared semantic model (Get data > Power BI semantic models, or the OneLake catalog) creates a live connection: your report uses the remote model's tables, relationships, measures, and security without copying data. You can add report-level measures but not Power Query steps or calculated columns. This is the thin-report pattern from 1.1.2, and you need at least Build permission on the model.
When an analyst must extend a shared model with their own data, they select Make changes to this model (or connect using DirectQuery for Power BI semantic models). This converts the live connection into a composite model: the remote model's tables are queried in DirectQuery mode, while new tables, such as an imported Excel targets file, sit alongside, and you can create relationships between them. The shared model remains unchanged for everyone else. The same pattern applies to Analysis Services sources.
⚠️ Exam Trap: A live connection can't add an Excel table or a calculated column. When the requirement is "use the certified sales model and add a local targets table related to Date," the answer is converting to a composite model (DirectQuery for Power BI semantic models). Importing the shared model's data or asking the owner to add the table both fail the "minimize effort, keep one source of truth" qualifier.
Reflection Question: Why does Power BI require Build permission on a semantic model before you can create a new report on it, when Read permission already lets you view reports built on that model?
