Create ClickHouse tables for a representative workload and choose sort keys and partitioning from actual query patterns. Load enough rows to reveal compression and merge behavior, then run common aggregates and a selective filter. Compare result correctness and p95 latency with the current store.
Test schema evolution, replication or restore, and user permissions. The right design depends on how data arrives and is queried; copying a demo table can hide expensive merges or unusable filters.
Akousa