Stark Informatics
Home · Solutions · Power BI

Power BI

The Microsoft Fabric presentation layer. Semantic models with DAX, Direct Lake mode over OneLake, reports, dashboards, paginated reports, and deployment pipelines that promote work through dev/test/prod.

GAWorkload · Power BI· 10 min read

What it is

Power BI inside Fabric is what you remember plus a great deal more. Reports and dashboards are still authored in Power BI Desktop or the web designer. Semantic models still use DAX. What changed: the semantic model sits on top of OneLake in Direct Lake mode (zero-copy, near-real-time), capacities are unified with the rest of Fabric, and deployment pipelines work across Fabric items (not just BI artifacts).

Storage modes & Direct Lake

Three storage modes still apply: Import (data copied into VertiPaq), DirectQuery (live SQL against source), and the Fabric-era addition — Direct Lake. Direct Lake reads Delta-Parquet files in OneLake column-by-column on demand, behaves like Import for performance, and refreshes the moment the upstream Lakehouse or Warehouse changes. Most production Fabric work belongs in Direct Lake.

i
Rule of thumb: Direct Lake by default. Import only when you need historical snapshots that the source doesn't preserve, or when you need calculated tables that Direct Lake can't yet support.

Reports

A report is the interactive analytical surface — visuals, slicers, drillthrough, bookmarks. In Fabric, reports live in workspaces, source from a semantic model, and promote through deployment pipelines. Best-practice reports follow a few rules:

  • One model, many reports — reports never re-implement business logic.
  • Bookmarks for navigation, not for state management.
  • Performance budget per page: under 5 seconds initial render on average.
  • Accessibility: alt text, tab order, sufficient color contrast, and keyboard-navigable visuals.

Dashboards

A dashboard pins tiles from one or more reports for an at-a-glance, monitoring-oriented surface. In Fabric, dashboards still serve their purpose, but the heavy lifting has moved to Real-Time Dashboards (for streaming) and to bookmarks within reports (for state-heavy interactive exploration). Use dashboards for executive snapshots and for KPIs that should remain visible in Teams.

Deployment pipelines

Power BI's deployment pipelines now extend across Fabric items: notebooks, semantic models, reports, and dataflows can move together from dev → test → prod with parameterized sources. Combine with Git integration for full source control. We treat deployment pipelines as production infrastructure — not as a Power BI feature.

Best practices

  • Treat the semantic model as the product. Reports are the marketing material; the model is the asset.
  • Direct Lake first. Import only with a documented reason.
  • RLS on the model, not in reports. See the RLS/OLS Pattern Pack.
  • Tabular Editor + Best Practice Analyzer. Run BPA in CI before publish. Adopt a custom ruleset.
  • Capacity planning. Watch p95 query duration in the Capacity Metrics report. CU-spike incidents are usually a single missing measure.

Common pitfalls

!
Re-implementing logic in DAX measures across multiple models. Build it once, reuse via shared semantic models or composite models.
!
Treating Direct Lake as "Import that auto-refreshes." It's column-store on demand. Calculation patterns that work in Import may need rewrites for Direct Lake's behavior with calculation groups and field parameters.
Semantic Models Lakehouse Warehouse Data Agents Real-Time Dashboards

Modernizing a Power BI estate?

Most estates we audit are 2× more expensive and 5× messier than they need to be. We'll tell you where the wins are.

See modernization service BI Governance Pack