Azure SQL Database expert — short-term advisory on clustered columnstore

We're a small US-based B2B SaaS company building a multi-tenant platform for in-building wireless / RF analytics (drive-test and DAS measurement data). Our measurement store is Azure SQL Database using a clustered columnstore index (CCI) with partitioning, on a vCore Business Critical tier, with read-committed snapshot isolation on. Backend is Python 3.12 / FastAPI talking to the DB via SQLAlchemy + pyodbc (ODBC Driver 18), authenticating through an Azure managed identity. We've just completed an internal CCI benchmark (compression, query latency, and bulk-load throughput on a ~7.5M-row corpus) and have a handful of interpretation-level decisions where we want an experienced second opinion specifically on Azure SQL Database (PaaS) behavior — not generic SQL Server. This is advisory: reviewing our benchmark findings and answering pointed questions, not writing production code. Initial scope (est. 3 hours, async + one or two calls): Sanity-check our warm-vs-cold columnstore query results and help us reason about cold-start behavior on Azure SQL DB, given that DBCC DROPCLEANBUFFERS isn't available on PaaS. How should we honestly characterize and mitigate a ~3s cold first-touch on a columnstore heatmap scan? Advise on service-tier choice: is Business Critical the right steady-state for a small-customer-base workload, or is General Purpose (or a smaller BC) defensible? How would each affect cold columnstore reads and cost? Review our bulk-load approach: pyodbc fast_executemany (we measured ~53× speedup), batch/commit strategy, and the DECIMAL-binding correctness concern under Driver 18. Are there Azure-native load paths we should prefer given managed-identity auth constraints (which ruled out bcp for us)? Comment on CCI ordering / partitioning choices for compression and segment elimination on our query patterns. If the fit is good, we'd like to keep you on a light ongoing retainer as our escalation path for Azure-SQL-specific questions (a few hours a month, on-call). You're a strong fit if you've operated clustered columnstore indexes at scale on Azure SQL Database specifically (not just SQL Server on-prem or on a VM), understand the PaaS differences in tiering / IO / diagnostics, and can explain tradeoffs clearly to a technical-but-not-DBA audience. We value people who say "it depends, and here's on what" over confident one-liners.