Placing filters near the source reduces data volume early in the pipeline, minimizing unnecessary processing and memory usage in downstream transformations. Filtering downstream means carrying irrelevant rows through multiple transformations, which significantly degrades mapping performance.