Design an ad content delivery system with A/B testing

Design an **ad content delivery (ad serving) system** that selects and serves ads to users, and supports **A/B testing / experimentation**. ## Core scenario A client (web/mobile app) requests an ad for a placement (e.g., homepage banner). The system must choose an eligible ad and return creative metadata quickly. ## Requirements ### Functional - Create/manage campaigns and creatives (basic campaign metadata is enough). - Targeting (at least: country/device/app placement; optionally user segments). - Select an ad to serve for each request. - Impression and click tracking. - Basic delivery controls (at least one): pacing/budget, frequency capping, deduplication. - **A/B testing**: run experiments that change selection/ranking logic or creatives; assign users deterministically to variants; log outcomes. ### Non-functional - Very low latency for ad requests (typically tens of milliseconds). - High throughput (can assume millions of requests/day+). - High availability. - Accurate enough reporting (define what must be real-time vs eventually consistent). ## Deliverables - High-level architecture and key services. - Data model/storage choices for campaigns, targeting, and events. - Serving path (request → decision → response) including caching. - Event logging pipeline and metrics. - A/B test design: assignment, exposure logging, metric computation, and guardrails. - Key trade-offs and failure modes.