The Analytics Tool Stack

• Name the layers of a modern analytics stack
• Match tools to the work they do (query, transform, visualize, explore)
• Choose the right tool for a task instead of the favorite one

• Knowing the stack tells you where a number comes from and where it can break.
• The real power of modern analytics lives in SQL — BI tools just render it.
• Choosing the right tool per task (BI for monitoring, notebook for discovery) cuts wasted effort.
• A coherent stack prevents spreadsheet kingdoms that rot alongside the real system.

Refrigerators (warehouse) hold ingredients, prep cooks (transformation) cut them, and the pass (BI) plates finished dishes. Notebooks are the chef tasting early prototypes — great for experimentation, not for feeding 500.

• The warehouse stores modeled, governed data and answers SQL reliably.
• Transformation tools (dbt, SQL) turn raw layers into metrics-ready models with documented definitions.
• BI tools (Power BI, Tableau, Looker) query the models and render Monitoring dashboards.
• Notebooks and Python add statistics, ad-hoc exploration and modeling for one-off decisions.
• Governance (lineage, access, definitions) binds the layers so numbers mean the same thing.

• A product analyst writes the canonical funnel in dbt, then builds a weekly dashboard in Tableau on top.
• A data scientist explores a churn question in a notebook, then ships the definition into the warehouse.
• An operator runs ad-hoc SQL directly on the warehouse for a one-time board answer.

• Keep analysis code in version control, even one-liner SQL used twice.
• Use BI for monitoring and notebooks for discovery — don't jam either into the other's job.
• Model definitions once in the transformation layer; never duplicate formulas in BI.
• Learn SQL before reaching for GUI tools; SQL transfers everywhere.

• Performing analysis in Excel by downloading warehouse exports — breaks at any scale.
• Duplicating the same metric definition across five dashboards.
• Buying a new BI tool to cure a discipline problem that is really a definitions problem.
• Letting notebooks grow into production pipelines nobody can run.

• Trace a dashboard number: BI → model → raw source. The bug is in one of the three hops.
• If SQL is slow, check loading, then indexes, then whether the BI tool is scanning more than needed.
• When two tools disagree on the same label, diff their formulas and filters.
• Use lineage maps in the transformation layer to find every consumer of a changed column.