Data and AI · 14 October 2026
Choosing a data platform without a religious war
Snowflake, Databricks, BigQuery, and Fabric can all work. The failure mode is picking from a slogan and discovering the operating cost later.
Start from the jobs: finance reporting, a product event stream, a data science team, a latency you can name. A platform that is perfect for notebooks and awkward for the CFO’s pack will lose the political fight.
Cost is a design choice. Warehouses get expensive when every team lands raw data and no one models it. Lakehouses get expensive when nobody owns tables. The platform does not fix ownership.
We shortlist two, run the same two pipelines on both if the choice is close, and write down the operating skills you will need on staff. A platform your team cannot run is a subscription to us, which is a bad reason to choose it.
Governance is not a phase two. Access and a handful of tested metrics belong in the first release, or the second team will build a shadow extract and you are back where you started.