VertiPaq compresses repeated values. High-cardinality strings (GUIDs, free-text comments) crush compression. Open VertiPaq Analyzer (or DAX Studio Metrics) and sort tables by size before blaming licensing.
Cuts that move the needle
- Delete columns not referenced by any measure/visual — especially wide VARCHAR dumps from ERP “just in case.”
- Split DateTime into Date + Time (or drop Time) so the date dictionary stays small; mark a proper Date table with Mark as date table.
- Replace text status codes with integer keys + dimension attributes; keep display names on the dimension, not the fact.
Relationship hygiene
Prefer single-direction many-to-one star schemas. Bidirectional “to make the slicer work” often forces larger query plans and hides model errors. Avoid both-direction on large facts. If a flat Excel export is the source, shape to star in Power Query (or warehouse) before Import — a 40-column denormalized fact rarely compresses or filters cleanly.
Re-measure after each cut: Dataset size in the service and local PBIX file size. If Cardinality of a column stays near row count, that column is not dimensional — reconsider keeping it in Import at all.