Topic 6 of 6
Data models and visualizations in Power BI
A good report starts with a good model: well-related tables, measures, and hierarchies. Then you choose the chart that tells the story well. The exam asks about the features of Power BI models and which visualization fits each need.
Contents
What is a semantic model?
Analytical models, called semantic models in Microsoft Fabric and Power BI, structure data for analysis. They're built from related tables and define:
Measures
The numeric values you want to analyze or report, such as revenue or quantity. In Power BI they're written in DAX (Data Analysis Expressions).
Dimensions
The entities you aggregate measures by: products, customers, time.
That way you can get, for example, total revenue by customer or items sold by product per month. In a Power BI model, aggregations are calculated at query time.
Fact and dimension tables
| Dimension table | Fact table | |
|---|---|---|
| Represents | Entities to group or filter by | Recorded events, such as a sale |
| Contains | A unique key per row and attributes (name, category, city) | The numeric measures (quantity, revenue) |
| Example | Product, Customer, Time | Sales |
- Most models include a Time dimension to aggregate across periods.
- A fact table related to dimensions forms a star schema; if dimensions relate to more tables (Category to Product), it's a snowflake schema.
- In Import mode, Power BI stores the data in an efficient, in-memory columnar store using the VertiPaq engine.
Attribute hierarchies
Hierarchies let you drill down or drill up through different levels of a dimension:
Product
Category → product
Customer
City → customer
Time
Year → month → day
When you view sales by year and drill down to the monthly breakdown, the model computes the aggregates for each level at query time. In Power BI Desktop's Model view you create relationships, define hierarchies, and set data types and formats.
Relationships and cardinality
Relationships connect the model's tables so dimension filters propagate to facts. Each relationship has a cardinality: the "one" side contains unique values and the "many" side can contain duplicates.
| Cardinality | When it appears |
|---|---|
| One-to-many (1:*) / many-to-one | The most common: from Product (one) to Sales (many) |
| One-to-one (1:1) | Uncommon; often a sign of a suboptimal design with redundant data |
| Many-to-many (*:*) | Infrequent, for complex requirements, such as sales targets at a different level of detail |
Microsoft recommends applying star schema design principles so dimension filters propagate efficiently to fact tables.
Storage modes
| Mode | How it works |
|---|---|
| Import | Stores a snapshot of the data in the model so visuals load quickly; you refresh to see new data |
| DirectQuery | Queries the source when visuals load; data isn't stored in the model |
| Direct Lake | Reads OneLake Delta tables directly into memory, without copying them into the model |
| Dual and Hybrid | Combine behaviors; a model with tables in several modes is a composite model |
Choosing the right visualization
Power BI includes many built-in visualizations, extensible with custom and third-party visuals. The most common:
Tables and text
The simplest. Tables show many related values; cards highlight important figures or metrics.
Bar and column
Visually compare numeric values for discrete categories.
Line
Compare values and, above all, examine trends, often over time.
Pie
Compare values as proportions of a total.
Scatter
Compare two numeric measures and spot a relationship or correlation.
Maps
Compare values for different geographic areas or locations.
Interactive reports
In a Power BI report, visuals for related data are automatically linked. Selecting a category in one visual filters and highlights that category in the other related visuals: for example, choosing the city of Seattle in a column chart makes the rest of the report show only Seattle's values.
AI-powered visualizations
Smart narratives
Written summaries that describe what a visual shows and update as data changes.
Copilot
Users ask natural language questions about the report and get summaries and visuals. It's the recommended replacement for the Q&A visual, which retires in February 2027.
Key influencers
Identifies which factors most strongly drive a metric and shows it in an interactive visual.
Decomposition tree
Drills interactively across multiple dimensions to see what contributes to a value.
Exam scenarios
| Need | Visualization | Why |
|---|---|---|
| Compare sales across product categories | Bar or column chart | Compares values for discrete categories. |
| See how revenue evolves month by month | Line chart | Shows trends, often over time. |
| Show what share of the total each region contributes | Pie chart | Proportions of a total. |
| See whether ad spend relates to sales | Scatter plot | A relationship or correlation between two measures. |
| Compare sales by country | Map | Values by geographic area. |
| Show one key figure, like total revenue | Card (text) | Highlights important metrics. |
| Find out which factors most influence customer satisfaction | Key influencers | Identifies a metric's drivers. |
| Drill across several dimensions to see what contributes to a value | Decomposition tree | Interactive breakdown. |
Common exam traps
❓ Are numeric measures stored in dimension tables?
No. They're in fact tables; dimensions hold the attributes to group by.
❓ What are hierarchies for?
To drill down or up between levels of a dimension, such as year, month, and day.
❓ What is the most common relationship cardinality?
One-to-many (or many-to-one).
❓ Which chart shows the correlation between two measures?
A scatter plot.
❓ And to show proportions of a total?
A pie chart.
❓ In DirectQuery mode, is data stored in the model?
No. It's queried at the source when visuals load; Import does store a snapshot.
❓ What replaces the Q&A visual?
Copilot, the recommended replacement; the Q&A visual retires in February 2027.
Did you get this topic?
Test what you just learned
An analyst wants to show how the company's revenue has evolved over the last 24 months. Which visualization is most suitable?
Ready to test "Power BI models and visualizations"?
Original questions, each with a full explanation.
Start a mock examSign in to save your progress.
You finished the last topic of this deep dive.
Back to the index →