Design a Real-Time Anomaly Detection System for E-commerce Events

This tests your ability to design a real-time data pipeline and apply ML to a business problem, accounting for seasonality. A strong answer outlines event ingestion (Kinesis), transformation (Lambda), and storage (S3). For the algorithm, reject static thresholds and propose a model that learns historical patterns (like AWS Lookout for Metrics) to define a 'drop' relative to expected traffic. The main red flag is suggesting a naive algorithm like 'a 20% drop from the last hour's average,' which ignores traffic cycles.