What is a Data Transformation?

This is the "T" in ETL and covers everything from trivial reformatting (converting a date string's format) to substantial restructuring (flattening a nested JSON object into relational rows, or aggregating thousands of transaction records into a single daily summary). Unlike normalization, which is about consistency, transformation is fundamentally about reshaping data to match what the next system or query needs. Transformation logic is one of the most common places for silent bugs to live, because a transformation that runs without error can still produce subtly wrong output - a currency conversion using a stale exchange rate, or an aggregation that double-counts because of a bad join. This is why transformation steps in a mature pipeline are unit-tested against known input/output pairs, not just trusted to run. AI has added a new transformation mode beyond rule-based scripts: an LLM can transform unstructured text into structured fields (summarizing a document into key-value pairs, for instance) in ways a deterministic script can't easily replicate.