Forum Discussion

D_PBI's avatar
D_PBI
Post Partisan
2 months ago
Solved

Training course recommendation on learning Natural Language to SQL within SQL Server?

Hi (I hope this is a suitable forum), We currently get our insights through Power BI (PBI) accessing data within SQL Server. Producing thsese reports/insights is labour intensive. With the introduct...
General Question
  • v-kpoloju-msft's avatar
    v-kpoloju-msft
    2 months ago

    Hi D_PBI, 
    Thank you for reaching out to the Microsoft Fabric Community Forum.

    Natural Language-to-SQL (NL2SQL) is possible within the Microsoft ecosystem using Azure OpenAI, Copilot Studio, Microsoft Fabric, and Power BI Copilot. The quality of results depends on having a well-structured SQL Server data model with meaningful table and column names, clear relationships, documented business definitions, and appropriate security controls. AI can enable users to ask questions in plain English and generate SQL queries, but it is generally best used alongside Power BI rather than as a complete replacement, since Power BI continues to provide governed reporting, dashboards, KPIs, and visual analytics.

    Since you are new to AI, a good starting point would be Microsoft's AI-900 Azure AI Fundamentals training, followed by learning about Azure OpenAI, Copilot Studio, and Fabric Copilot capabilities. I would also recommend building a small proof of concept using a subset of your SQL Server data to evaluate answer quality, security requirements, and business adoption before considering a broader implementation. This approach will help determine whether Natural Language querying can effectively support your reporting and analytics needs.

    Refer these links:

    1. https://learn.microsoft.com/en-us/fabric/fundamentals/copilot-fabric-overview? 
    2. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-introduction? 

    Hope this helps. If you have any questions regarding this, please feel free to reach out to us. We will be happy to help.

LiamOPax8's avatar
LiamOPax8
New Member
2 months ago

I’m actually looking into this myself at the moment, so I’m not an expert yet, but I’ve been testing a few approaches in the Microsoft stack.

From what I’ve seen so far, yes, natural language querying over SQL data is possible, but it does not completely remove the need for something like Power BI or a well-structured data model. Your instinct is probably right. The clearer and more descriptive the data is, the better the results tend to be.

If you’re already using SQL Server, Azure, and Power BI, I would stay within that ecosystem and look at things like Power BI, Microsoft Fabric, and Copilot, rather than introducing lots of separate tools.

One thing I’ve been experimenting with is using Fabric to bring data into a lakehouse, then using Copilot in Notebooks to explore the data in natural language. This is not a full end-user solution, but more a way to understand how well AI can interpret the dataset.

For example, you can give it a prompt like this:

 
Analyse all tables in the attached lakehouse.

 

For each table:
- Describe what the table likely represents
- Classify it as one of:
  - transactional (fact-like)
  - reference (dimension-like)
  - bridge/helper
  - unknown

 

Also include:
- row count estimate
- key-looking columns
- any obvious issues (nulls, duplicates, naming issues)
``
 

That will not replace reporting on its own, but it is useful for:

  • understanding how AI-ready your data is
  • seeing how well natural language works against your schema
  • identifying gaps in naming, structure, and relationships

From there, you can move into more user-style questions like:

  • “What are the top customers by revenue this quarter?”
  • “Show trends over time”

If you are starting from scratch, I would focus training on:

  • data modelling
  • Power BI and semantic models
  • Microsoft Fabric
  • Copilot and natural language features on top

So overall, it is achievable, but it works best as an addition to a well-modelled dataset rather than a full replacement for BI.

Would be interested to see what others are using as well.