Tableau Data Sources to Power BI Semantic Models: Embedded, Shared, and Endorsed

Two things trip people up here, and only one of them is technical. The vocabulary changed: what Power BI called a dataset is now called a semantic model, and Microsoft's documentation uses the terms model and semantic model interchangeably. The substantive question is the one Tableau made you answer too, just under different names: does this data live inside a single report, or does it become a shared asset that many reports connect to?

Tableau data sources in Power BI terms

The decision the migration forces

In Tableau, embedding a data source in a workbook is the path of least resistance and publishing one is a deliberate act. Power BI works the same way, and a lift-and-shift migration inherits whatever the Tableau estate happened to do. That is the moment to choose rather than inherit.

Publish a shared model when several reports answer questions from the same numbers, when the measure definitions are the thing that must not diverge, or when report authors should never hold source credentials. Keep the model inside the report when one report owns its data end to end, when the refresh cadence is specific to it, or when a shared model would collect fields nobody else uses. A workbook estate with thirty embedded copies of the same sales data is telling you which way to go.

Endorsement replaces the naming convention

Every Tableau estate has an informal signal for which data source to trust: a prefix, a folder, a wiki page. Power BI has endorsement built in, at two levels with different rules. Promotion can be applied by any content owner or workspace member with write permissions, and simply says the content is worth using. Certification means the content meets the organisation's quality standards and can only be applied by reviewers the Power BI administrator has authorised.

Endorsed semantic models can also be made discoverable, so users without access can find them and request it rather than rebuilding their own copy. That is worth configuring during migration rather than after, because the alternative is exactly the sprawl the migration was meant to end. Write the description while you are there: it is what users see in the semantic models hub.

What a semantic model actually is

Technically, a Power BI Desktop model is an Analysis Services tabular model, and it comes in three modes: Import, DirectQuery and Composite. That has two consequences worth knowing during a migration. Import models must be fully loaded into memory to be queried and need refresh to stay current; DirectQuery models need live connectivity to the source, and a gateway when that source is not reachable over the internet.

It also means the model is not a per-report convenience the way a Tableau embedded data source can be. Relationships, measures, formatting, hierarchies and row-level security all live in it, and they are exactly what is worth writing once instead of thirty times.

Row-level security and ownership

Tableau user filters become RLS roles, which are dynamic or static: dynamic roles filter by the report user, static roles apply the same filter to everyone assigned to the role. Both are defined in the model rather than per report, so a shared model gets its security defined once, which is another argument for publishing.

Ownership is the operational detail people meet later. If you are not the semantic model owner, its settings are read-only until you take ownership, and there is an API for taking over when an owner leaves the organisation. Decide who owns each model during the migration rather than discovering it during a failed refresh.

What Antares does with data sources

The Analyzer inventories every data source in every workbook, records whether each is live or extract-backed, and captures what it connects to. Read across an estate, that is the evidence for the decision above: which connections repeat, which are unique, and therefore which become shared semantic models rather than thirty private copies.

The conversion itself produces the report layer, and analysis reads metadata only, never business data. Model ownership, endorsement and security stay where they belong, with the people who run the platform. Run the free Analyzer to see how the connections repeat across your own workbooks.

Related reading: extracts to Power BI dataflows and blending to model relationships. Primary sources: Tableau's published data source documentation and Microsoft's guide to connecting reports to shared semantic models.