Most Power BI performance problems are diagnosed in the semantic model, but are caused long before it exists.

With Microsoft Fabric, transformations, shaping, and architectural decisions increasingly move upstream into Dataflows Gen2, SQL endpoints, and notebooks. These early choices silently determine model size, refresh times, DAX complexity, and whether Direct Lake succeeds or fails.

In this session, we explore how upstream design decisions around grain, shaping, aggregation, and transformation impact downstream semantic models. Through real-world patterns and architectural examples, you’ll learn when logic belongs upstream, when it belongs in the model, and why “fixing it in DAX” is often a sign of deeper issues.

The goal is not to eliminate DAX—but to let it do what it does best, on top of data that was prepared right.

This is an advanced technical session intended for experienced Power BI and Microsoft Fabric practitioners, including senior analysts, analytics engineers, and BI architects. Attendees are expected to be familiar with semantic modeling concepts, star schemas, and basic DAX.

The session focuses on architectural and performance trade-offs across Microsoft Fabric components such as Dataflows Gen2, SQL endpoints, notebooks, and semantic models, with an emphasis on enterprise-scale design decisions.

Preferred session duration is 50 minutes (approximately 45 minutes of content plus Q&A). The session is presentation-driven and may include prepared examples or architectural walkthroughs, but does not require live coding or audience participation.