SQLBits 2025

Earlier this summer, four of us from Lytix had the opportunity to attend SQLBits 2025, one of the largest data platform conferences in Europe. Held over four packed days, the event brought together data professionals, developers, and thought leaders from across the globe. 

Our goal? To learn from the best, explore the latest in data and AI, and bring back fresh insights to strengthen our work at Lytix.  

Since we are all about knowledge sharing at Lytix, we summarized our favorite sessions for you below. Enjoy the read! 

Simplify your DAX with window functions 

Window functions are a recent addition to DAX that allow you to both simplify and optimize your code. In his session, Greg Strzyminski introduced the main functions: INDEX, OFFSET, and WINDOW, and showed how they handle typical use cases like referencing previous or next items, calculating running averages, and building cumulative sums. These functions not only make your code easier to write and maintain but are also more performant compared to traditional DAX patterns, so go ahead and start rewriting your old measures . They also lay the foundation for the upcoming visual calculations in Power BI. While the syntax might look a bit different at first, it’s surprisingly approachable once you get the hang of it. With clear examples and practical guidance, this session made a strong case for making window functions part of your day-to-day DAX toolbox. 

Deep Dive into Power BI Direct Lake 

Direct Lake is a Power BI storage mode that speeds up data access by loading Delta tables from OneLake into memory for fast analysis., but getting the most out of it means understanding how it works behind the scenes. In his session, Patrick Leblanc (yes, that guy from Guy in a Cube) explained how Direct Lake achieves its speed and what you need to set it up correctly. First off, it’s important to know that Direct Lake only works with Microsoft Fabric F SKUs and isn’t supported on Power BI Pro, Premium Per User, or Embedded SKUs. 

A key factor behind Direct Lake’s improved performance is Microsoft’s proprietary V-order format for writing parquet files.

While Direct Lake works with any Delta table, it performs noticeably better with V-ordered tables. V-order uses the same compression algorithms as Power BI Import models, enabling more efficient data reads. Plus, V-ordered Delta tables remain fully compatible with any tool that supports Delta.

Patrick also demonstrated how to interpret Delta Analyzer logs and monitor performance end-to-end, essential knowledge if you’re building for scale in Fabric. 

Data streaming in SQL 2025 

Microsoft has launched a new capability in Azure SQL Database called Change Event Streaming (CES), now in public preview, which makes it much easier to build real-time data pipelines with SQL.  

With CES, changes in your SQL database (INSERT, UPDATE, and DELETE operations) are streamed directly to Azure Event Hubs in near real-time by reading the transaction log. Unlike older approaches like Change Data Capture (CDC), CES sends only committed transactions as individual events, minimizing latency and eliminating double logging overhead.  

While it’s currently limited to Event Hubs as a destination and currently there is no built-in row-level or column-level filtering, it’s designed for modern streaming use cases like pushing SQL data to Kafka consumers, Spark jobs, or Fabric Lakehouses. CES represents a shift from traditional SQL-to-SQL replication toward integration with event-driven architectures, offering a lightweight, high-speed alternative for getting data out of your database and into your cloud analytics stack. 

Copilot in Microsoft Fabric 

When planning to use Copilot in Microsoft Fabric, there are a few important things to know around capacity, consumption, and monitoring that users need to understand.  

1. Copilot capacity is available starting with F2 SKUs 

2. Capacity region matters for Copilot, not your tenant region. Enabling Copilot is an all-or-nothing decision for your capacity. You can’t selectively enable it for specific workspaces or locations.  

3. Data handling: conversation history for notebooks can be stored outside your capacity region, unlike Power BI, where history stays within region. As for what data Copilot actually uses. “It depends”. It depends on the prompt, query results may be sent as part of processing, but they aren’t stored afterward.  

4. Capacity consumption is tricky to estimate because it varies by model, but a general rule is 100 CU seconds for input prompts and 400 CU seconds for output. But make sure to test your real-world scenario. 

Although we see that Copilot requests often use less than 0.1% of an F64 capacity, there may still be additional model-processing overhead which are hard to monitor. Monitoring tools are (currently, july 2025) on the roadmap. Today, usage is tracked in the capacity monitoring app (under “Copilot in Fabric”).  

Asking questions to get meaningful answers 

In the age of AI, where almost all information is available with the click of a button, it takes courage to say, “I don’t understand, can you explain that to me?” Asking these kinds of questions spurs learning, stimulates discussion, and fuels innovation. During his session, Benni De Jagere laid out a framework for how to ask questions that lead to meaningful answers.  

The main components of the framework are: 

• Inform yourself before asking a question. Go and look up information on your question. These days there is already a ton of information online. 

• Provide context. A question without context is meaningless.  

• R-E-S-P-E-C-T. Respect the person that you are asking a question to. It is, for now at least, a person of flesh and blood. Treat them like it.  

Although the framework was intended for human interaction, it’s equally applicable when engaging with AI. Learning how to properly formulate a question is an incredibly valuable skill, so time to start practising. 

Dataverse in connection with Power BI and Microsoft Fabric 

In Microsoft Fabric everything is built on top of OneLake. 
A unified data storage for anything and everything in fabric. This makes our lives much easier. 
Now, we also know every workspace is practically a server with its own databases underneath. 
Onelake gives us the option to easily create a shortcut from one of these workspaces to the other to access the data. 

When you try to create a shortcut to Dataverse it won’t work as it is not in the OneLake. 
To do this you will need to set up Fabric Link. 

These are the steps and Prerequisites you need for it to work 

• Fabric SKU/Premium per Capacity 

• Enable Tenant Setting: 

•  Users Can access data stored in OneLake with apps external to Fabric 

• Workspace Admin Role 

• System Admin Role 

• Same Geographic Locations 

Once set up, it will synchronize the data from the Dataverse into a delta log in Fabric in ‘Real Time’. Sadly it is not truly Real Time as there is a possible delay of up to 60 minutes. Next to this, it will take all the columns available in your Dataverse table. This means all the automatically generated columns created by the Dataverse will be synced as well. There can be a lot of them making it hard to find the data that you actually need. 

It is amazing that we can access data in a Dataverse with a shortcut. 
But it has a lot of prerequisites and caveats that we need to take into consideration, making it quite a hassle to get it set up and used. 

Automating Engineering with AI 

AI has been taking the world by storm. It has been appearing everywhere and is being used by everyone. Sometimes in a good way, often in a less then optimal way. But no matter what you think of AI, it is here to stay. So, we have two options. Embrace it or be left behind.  Embracing does not mean using it whenever we can and just copying and pasting the outcome. What it does mean is taking advantage of its capabilities, understanding it and using it mindfully. 

Currently there are already tools in the market that create ETL pipelines with an AI algorithm. 
Its infused with every data tool we can think of in every step of the way. It is inevitable in the data world to avoid it. 

In the end AI is not there to replace use. It’s here to augment our work and give us the opportunity to focus on the tasks that matter instead of documentation, optimizing, and so on. So don’t be afraid of using it, because if you don’t you will lose that competitive edge. 

Closing 

From hands-on sessions to inspiring keynotes and many hallway conversations, SQLBits 2025 delivered on every front. We gained new knowledge on the technical front but also walked away with valuable new perspectives on communication and leadership. This is also what makes SQLBits such a standout conference: the variety of topics and the welcoming community.  

Thank you for reading our little summary. If you want to have a chat on some of the topics, you know where to find us.  

Written by