This document summarizes audience targeting techniques in digital advertising. It discusses audience segments, which are named sets of device IDs or cookies. Segments can contain billions of IDs and there may be hundreds of thousands of segments. It then discusses using key-value storage and Bloom filters to store and query large audiences in real-time for ad targeting across multiple data centers while meeting latency and cost requirements. Bloom filters provide a probabilistic alternative to reduce costs compared to precise key-value lookups but cannot guarantee accuracy.