Get startedGet started for free

Semantic models in Fabric

1. Semantic models in Fabric

So far, you've loaded and cleaned data in lakehouses and warehouses. But report builders need more than tables: they need business-friendly fields and trusted metrics. A semantic model provides that layer.

2. A semantic model adds business meaning

Decision makers rarely read raw sales rows. They ask questions like "What's our revenue by product?" A semantic model sits between stored data and the reports that answer those questions. Its source item, such as a warehouse, supplies the data. Tables expose fields a report can use, relationships let a filter on one table reach another, and measures store reusable calculations. For example, a Sales table stores only product IDs, but a relationship to Products and a revenue measure let a report show revenue by product name.

3. DAX defines model calculations

Measures are written in Data Analysis Expressions, or DAX, the formula language of semantic models. A total revenue measure stores its logic once, so every report gets the same result. Later in this chapter, you'll read DAX and compare measures with calculated columns.

4. Creating a semantic model

A model gets its data from a supported Fabric item, such as a lakehouse or warehouse. The source owns the tables; the model selects some of them and adds names, relationships, and calculations as a separate layer, so editing the model never rewrites source rows. To create one, open the source item and select New semantic model on the ribbon. If it's grayed out, refresh the item so new tables register.

5. A query creates warehouse source tables

A source needs tables, such as lakehouse Delta tables or warehouse tables. In an empty warehouse, the T-SQL template opens the query editor shown here. This script has two statements. CREATE TABLE defines a Products table with a numeric ProductID and a text ProductName. INSERT INTO adds one row: product 101, Trail Backpack. In the exercises, a longer script also creates a related Sales table. The goal of this course is to understand what the script builds.

6. Semantic model editor: the ribbon

The new model opens in the semantic model editor. At the top, the ribbon holds commands to create measures, manage relationships, define security roles, and check model health.

7. Model canvas and Properties pane

The canvas shows each table in the model, and each table has a three-dot menu with more options. Relationships appear as lines between tables once you create them. Select a table or column, and the Properties pane shows its settings, while the Data pane lists every model object. If the Properties pane covers the canvas, minimize it.

8. View tabs and zoom

Along the bottom, view tabs switch between the Model diagram, DAX query view for testing DAX, and Tabular Model Definition Language, or TMDL, which shows the model as code. Zoom controls help you navigate larger models.

9. Composite models can mix table strategies

A composite model mixes tables in different storage modes, so a huge detail table and a small summary table can sit side by side. OrderLines holds 500 million rows, while MonthlySalesAgg holds 2,400 monthly totals by region. Each table has a grain: the level of detail one row represents. A monthly regional sales chart reads the aggregation because its grain matches. Coarser questions, like yearly totals by region, can also add up the monthly rows. A customer's order history still needs OrderLines, the only table with that detail. Routing each question to the smallest table that can answer it scans fewer rows.

10. Let's practice!

Let's practice!

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.