Advanced
Open
Pro
Design an anomaly detection pipeline for 10,000 business metrics with 10 alerts a day
Part of the ML System Design Interview path →
Part of the Advanced Data Science & Experimentation path →
Your company has 10,000 daily business metrics (orders per region, signups per channel, spend per product, and so on). Each has weekly seasonality, many have trends, some are intermittent, and a few have promotions or holiday effects. On-call can handle about 10 alerts a day in total. Design the anomaly detection pipeline.
Cover: the per-metric detector, how you turn the alert budget into thresholds, how you deal with level shifts, how you group alerts, and what you do for the metrics that cannot be modelled well.
Share this question