comScore is an internet analytics company that processes over 1.5 trillion digital interactions per month. They were tasked with calculating campaign metrics for over 130 billion records spanning 92 days. Their initial MapReduce approach did not scale due to large data shuffles. To improve performance, they partitioned and sorted the data by cookie daily before using a custom input format to merge partitions and do map-side aggregations, reducing shuffle sizes and allowing combiners to be used. This improved processing time from 35 hours to 3 hours without hardware changes.