The request

From Kevin Nguyen, Head of Engineering
Your goal
Match each of Kevin's three requests to the right way of connecting.
What Kevin gets, and why
The lesson
Before Power BI shows a single chart, you decide how it reaches the data. The choice decides how fresh the numbers are, how fast the report feels and which features you can use.
Where you choose
Home → Get data → SQL Server: under the server and database boxes, Data Connectivity mode offers Import or DirectQuery. Excel and CSV files are always imported. For a live connection you start somewhere else: Home → Get data → Power BI semantic models (or OneLake catalog), then pick a published model. After loading, the bottom-right corner of Desktop shows the storage mode, and in Model view each table’s Properties → Advanced → Storage mode tells you the same per table.
Import: a copy in the model
Power BI reads the data once and stores a compressed copy inside the .pbix (and, after
publishing, in the Service). Every visual is answered from that copy, in memory, so reports are
fast and every Power Query step and DAX function works. The price is freshness: the copy is as
current as the last refresh. In the Service you schedule it (Semantic model → Settings →
Refresh), up to 8 times a day on a Pro workspace and up to 48 on Premium capacity.
Import is the default and the right answer most of the time, including for Harbor & Pine’s 12,000 orders.
DirectQuery: the data stays in the source
Nothing is copied. Each time someone opens a page or clicks a slicer, Power BI sends queries to the database and draws what comes back. The numbers are as current as the database, and the model stays small however big the table is. The costs: a page is only as fast as the database answers, every reader adds load on that database, Power Query is limited to steps the source can run, and some DAX is slower or unavailable. Use it when the data is too big to copy or must be current within minutes.
Live connection: reuse a published model
A live connection connects a new report to a semantic model that already exists in the Service (or to Analysis Services). You get its tables, relationships and measures as they are, with no Power Query and no Table view; you can add report-level measures, but not change the model. It’s how a company keeps one certified definition of “revenue” while many people build reports on it. (With the right settings, Make changes to this model turns it into a composite model, which mixes the published model with your own tables.)
Choosing, in one table
| Import | DirectQuery | Live connection | |
|---|---|---|---|
| Data copied? | Yes | No | No |
| Freshness | Last refresh | Current | The published model’s |
| Speed | Fastest | Depends on the source | Depends on the model |
| Power Query, model edits | All | Limited | None |
About this level
The practice levels in this track work like Import: the store’s data is loaded into a model in your browser, and every measure is answered from it. This level is a quiz because there’s no database to connect to here.
Common mistakes
- DirectQuery by reflex. Choose it for size or freshness, not “just in case”.
- Importing a copy of someone else’s model. Connect live to it; two copies drift apart.
- Forgetting the refresh. An imported report without a schedule stops at publish day.
Lantern lit!
Practised: import modedirectquerylive connectionstorage moderefreshsemantic model
- Solved
- No hints
- First run
