Back to All Engineering Notes
Semantic Models•10 min read•2026-10-11

Incremental Refresh in Power BI: Setup, Pro Licence Limits and Common Failures

Incremental refresh works on Pro. Here is how to set it up correctly, what you only get with Premium or Fabric, and why most failed setups come down to query folding.

Core Engineering Takeaways

  • •Incremental refresh is supported on Pro, PPU, Premium and Embedded. Only the real-time DirectQuery partition needs Premium, PPU or Embedded.
  • •It needs two date/time parameters named exactly RangeStart and RangeEnd, used to filter the table in Power Query.
  • •The filter must fold to the source; otherwise Power BI loads all the data and filters it afterwards, and refresh gets no faster.
  • •After publishing, you cannot download the model back as a PBIX, and republishing from Desktop replaces the partitions.

1. What incremental refresh does

Without incremental refresh, every scheduled refresh reloads the whole table. With it, Power BI splits the table into date-based partitions, keeps the historical ones, and refreshes only the recent ones — for example store 5 years, refresh the last 10 days.

The benefits are faster, more reliable refreshes, less load on the source system and less memory used during refresh. On Pro, where a scheduled refresh has a two-hour time limit, it is often what keeps a growing model refreshable at all.

2. What works on Pro — and what needs Premium or Fabric

A common misconception is that incremental refresh needs Premium. According to Microsoft's documentation it is supported for Pro, Premium Per User, Premium and Embedded models.

What capacity or PPU adds:

  • Real-time data with DirectQuery — a hybrid table whose latest partition queries the source live. This needs Premium, PPU or Embedded (and Fabric capacity).
  • The XMLA endpoint, which lets you refresh individual partitions, inspect them in SSMS or Tabular Editor, and run advanced refresh scenarios.
  • Bigger models. On Pro the 1 GB model limit still applies to the whole table — incremental refresh reduces refresh time, not model size.

3. Setting it up step by step

  1. In Power Query, create two parameters named exactly RangeStart and RangeEnd (case-sensitive), type Date/Time, with any sample values.
  2. Filter the fact table's date column using both parameters — one boundary inclusive, the other exclusive, so no row lands in two partitions.
  3. Check that the filter step folds: right-click the step and look for View Native Query, or check the query plan.
  4. Close and apply. In the table's Incremental refresh settings, set how much history to archive and how much to refresh (for example 5 years and 10 days).
  5. Optionally enable Only refresh complete days and Detect data changes (below).
  6. Publish and run the first refresh in the service. It loads all history, so it takes longest.
Using >= on one side and < on the other prevents duplicate rows at partition boundaries.
// Filter step — RangeStart inclusive, RangeEnd exclusive
= Table.SelectRows(
    Source,
    each [OrderDateTime] >= RangeStart and [OrderDateTime] < RangeEnd
)

4. Detect data changes and complete days

  • Detect data changes uses a date/time column such as LastModified to refresh only the partitions in the refresh window whose data actually changed. It saves work when late-arriving changes are rare.
  • Only refresh complete days makes sure every refresh includes whole days, so a refresh at 10am does not load half of today. It is required if you add a real-time DirectQuery partition.

If your source date is an integer key (such as 20261011) instead of a date/time, convert RangeStart and RangeEnd to the same integer format inside the filter step with a small function — the parameters themselves must stay Date/Time.

5. Why incremental refresh fails (and how to fix it)

  • The filter does not fold. Steps such as adding an index column, merging with a non-folding query or certain type conversions before the filter stop query folding. Power BI then pulls every row and filters locally, so each partition loads the full table. Move the date filter as early as possible, directly after the source step.
  • The source cannot filter by date. Incremental refresh works best on relational sources. File-based sources can work if files are organised by date, but every partition may still read every file.
  • Duplicate rows at boundaries. Both conditions use >=/<=. Make one side exclusive.
  • Republishing wipes the partitions. Publishing the PBIX again replaces the model in the service, partitions and data included, and the next refresh reloads all history. For ongoing changes, deploy metadata-only through the XMLA endpoint or with Git and deployment pipelines.
  • No way back to Desktop. Once refreshed in the service, the model cannot be downloaded as a PBIX. Keep the source PBIX — or better, a PBIP project in Git.
  • First refresh times out. The initial load brings in all history. On capacity you can bootstrap partitions over XMLA; on Pro, start with a shorter archive window and extend it later.

Full details: Microsoft's incremental refresh overview.

Power BI Diagnostic Audit

Refreshes still too slow?

Check how your model and refresh schedule load a capacity, or get a fixed-scope Health Check of the model and its refresh design.

Get the next Engineering Note

Practical DAX, semantic model and Fabric notes. 1–2 a month, unsubscribe any time.

1–2 useful notes a month. Unsubscribe at any time.

Piyush JainPower BI & Fabric Engineer
#Incremental Refresh#Power Query#Refresh#Query Folding#Pro