Glossary · Analytics

Anomaly Detection in Metrics

Anomaly detection in metrics is the automated identification of unusual values in business data–spikes, drops, or drifts that deviate from normal patterns–so problems surface before someone happens to look at a dashboard.

Dashboards only catch what someone is watching. Anomaly detection watches everything: it learns a metric's normal range and rhythm–weekday versus weekend, seasonal cycles–and flags when reality leaves that band, like signups dropping 40% after a broken deploy or refunds tripling overnight. The business case is response time: the gap between "it happened" and "we noticed" is where revenue quietly leaks. Approaches range from simple thresholds (alert below 50 signups/day), which are predictable and easy to reason about, to statistical models that adapt to trends–and for most teams, well-chosen thresholds catch the incidents that matter.

Fig — every answer ships with the tables, rows and SQL behind it.