This post covers Databricks' Predictive Optimization (PO) in Unity Catalog, which became the default platform behavior in 2025 for autonomous lakehouse table maintenance.
- •PO automatically runs OPTIMIZE, VACUUM, CLUSTER BY, and ANALYZE based on observed query and write patterns, eliminating manual table tuning
- •Automatic Statistics (GA) delivered up to 22% faster queries using stats-on-write (7-10x more performant than ANALYZE TABLE) and background refresh
- •Optimized VACUUM using the Delta transaction log achieved up to 6x faster execution and 4x lower compute cost compared to standard directory listing approaches
- •Automatic Liquid Clustering (GA) analyzes query telemetry to auto-select optimal clustering keys, reducing data scanned with zero manual configuration
- •Upcoming 2026 features include Auto-TTL for automated time-to-live row deletion and a Data Governance Hub dashboard for PO observability and ROI tracking
Databricks