Data Drift and Concept Drift Detection — ML Ops & Data Science in Production

Machine learning models are trained on historical data and then deployed into a world that keeps changing. As the world evolves, the statistical properties of incoming data may shift away from what the model was trained on — a phenomenon called drift. Drift is one of the primary causes of silent model degradation in production: the model still runs, still returns predictions, but those predictions become increasingly wrong. Understanding the different types of drift, how to detect them statistically, and which algorithms to use for continuous monitoring is a foundational production ML skill that separates experimental models from resilient production systems. The two failure modes this lesson separates — the inputs changing versus the input-to-outcome relationship changing — demand different detectors and different responses, and conflating them is the most common monitoring mistake in practice. A model can be perfectly healthy under shifted inputs it has seen the like of before, and fatally broken on inputs that look identical to training data.

Analogy🏏Cricket

🏏 Think of it like cricket: Evidently AI is the IPL's official analytics platform — rather than each franchise building their own stats system, they use a shared platform that automatically computes every standardized metric: batting averages, economy rates, strike rates, net run rates. When Virat Kohli's performance drifts from his baseline, the platform highlights it automatically with charts. Evidently does the same for ML models: instead of each team coding their own drift detectors, they use Evidently's pre-built metrics and get standardized, comparable reports automatically. The standardization is the strategic point, not a convenience: because every franchise reads the same metric definitions, a drift score of 0.3 means the same thing in every dashboard, reports can be compared across teams and seasons, and a new analyst is productive on day one. Hand-rolled monitoring scripts fail exactly here — every team's 'drift check' quietly means something different, and nobody can audit whose alarm was right.