How to connect to the ADX raw data?

[francesco1119]

Hello @flanakin , I would like to help in the project.
I will start by proposing a few changes in the Power BI dashboards.

But how to connect to the ADX raw data?

If I try to connect to the ADX it asks for Import or DirectQuery:

And then if I do that I see Hub and Ingestion

• If I Direct Query them I have array problems (arraystring)
• If I Import... is that the real way to do it? Importing years of data? I don't think so, that would be too slow.

And if I connect using the Power BI Semantic Model

I could then import.
But if I take CostSummary I see Costs and PricingUnits

And all the columns looks completely different from the raw data that I'm supposed to see in the Parquet export.
These data are maybe already cleaned up and you are OK with me playing with those.
But we have to discuss this before I start working.

[DakotaB75]

Hi @francesco1119, great question...

Power BI is not expected to connect directly to the raw ADX ingestion tables.
Those tables are optimized for ingestion and traceability, not for analytics.

Intended flow

graph LR
A[Parquet exports] --> B[ADX raw tables]
B --> C[KQL transformations]
C --> D[Curated / semantic tables]
D --> E[Power BI Semantic Model]
E --> F[Dashboards]

Loading

Because raw tables contain arrays and nested structures, DirectQuery or importing years of data does not scale well.
The curated / semantic tables intentionally flatten and normalize the schema, which is why they differ from the Parquet exports.

If your goal is to improve dashboards, working on top of the semantic layer is the safest approach.
Happy to align — I’m currently looking into ADX → Power BI docs and dashboard improvements.

[francesco1119]

OK, it's all clear now.
I will just work on the CostSummary.kql.pbit that I can find online and propose to merge.

One last question: where can I find the KQL code?

[DakotaB75]

i @francesco1119, there isn’t a public repository with the semantic model KQL code for CostSummary that’s shared by Microsoft. If your project has one, it should be in your team’s internal repo (GitHub/DevOps, etc).

Alternatively, you can open the .PBIT/.PBIX file, go to Transform data → Advanced Editor and you’ll see the actual KQL query being used in each query step that’s often the easiest way to extract or review the KQL being run by Power BI.
Microsoft Fabric Comunidad

[RolandKrummenacher]

@francesco1119
Can you share a bit more detail about the arraystring errors you are seeing with DirectQuery?

Specifically:
• which table or column triggers the error
• whether the column is of type dynamic (array or object) in ADX
• and at what point Power BI raises the error (preview, visual, filter, relationship)

That would help clarify whether this is a schema related limitation or a specific query pattern issue.

[francesco1119]

Yeh, this happens when I target the Hub database:

The very first function fails and makes fail all the rest

[RolandKrummenacher]

The lower-case functions (such as "arraystring") are only helper functions (some string manipulation stuff). You don't need those. I suggest you start by referencing "Costs_v1_2" only.