Data Wrangling

We perform most data wrangling in Power Query.

Since M is not the easiest language to maintain and tool support in Power BI is limited, it is essential to keep the structure clean.

A solid Power Query/M setup focuses on:

Performance and Query Folding

Filter and reduce rows as early as possible so that connectors can fold filters back to the source and avoid loading unnecessary data.
Remove unused columns (use "remove other columns") before merges and heavy calculations to minimize data volume and lower the cost of joins.
Perform expensive operations (merges on large tables, complex calculated columns, grouping, cross-source joins) as late as possible to keep earlier steps foldable.

Set correct data types early, but preferably after row/column reduction, so profiling and previews are accurate without overloading the engine.
Apply text cleanup (trim, clean, case transforms) only on the columns actually needed to avoid breaking folding and wasting CPU.

Use a modular approach: split logic into staging queries (raw cleaned data), transformation queries, and final presentation queries, referencing rather than duplicating queries.
Give meaningful names to queries, steps, and columns; avoid default names like “Filtered Rows 2” so that M code is readable and maintainable.
Group related queries in folders (staging, dim, fact, parameters, lookup) so models stay understandable as they grow.
Avoid having transformations on your dim and facts, do the transformations rather in the source queries.

Use parameters for environment-dependent values (DSNs, server/database names, date ranges, company IDs) to make solutions portable between dev/test/prod.

Format and comment M code (in Advanced Editor) to aid future debugging, and consider external version control for more complex projects.