Data · 2026-01-22 · 10 min
A modern data stack without the sprawl: warehouse, contracts, and one dashboard
Most companies do not need five reverse-ETL tools. They need a warehouse, a few well-owned pipelines, and dashboards that match how decisions are actually made.
The modern data stack exploded into a catalog of point tools. Spend follows: ingestion, transformation, catalog, quality, reverse ETL, BI. For a 40-person company, that is usually too many contracts and too few owners.
A stack that fits a small data team
dbt’s research and countless internal post-mortems agree: unused dashboards rot trust. If a chart has not been opened in 90 days, archive it.
- One warehouse (BigQuery, Snowflake, or Redshift) as the system of record for analytics.
- EL with a managed loader; transform in SQL with versioned models.
- Data contracts on the five tables executives actually quote.
- One BI tool, with certified dashboards and a graveyard policy for unused ones.
AI on top of messy data is theatre
Do not bolt a LLM onto unmodeled event streams. Clean the grain of orders, users, and tickets first. Then retrieval and forecasting become boring — which is the point.
Fresh, contracted tables beat a vector database full of last year’s CSV dumps.