Home › Key concepts › Drift
Key concept · Run (advanced) · the same on every platform
Drift
The world changes and the model quietly gets worse.
What it is
Data drift is when the inputs change, such as new products or a new customer mix. Concept drift is when the relationship changes, such as what predicts churn after a price rise. You watch input distributions and live accuracy, and retrain when they move.
Why it matters
Models don't break loudly. They decay, and nobody notices until a business number drops.
An example
After a competitor launches, the share of price-sensitive customers jumps and the churn model's recall falls from 70% to 45%.
The common mistake
Deploying a model and never measuring it again.
Where you meet it on each platform
The idea stays the same; only the product name changes.
| Job | AWS | Azure | Google Cloud | Databricks | Open source |
|---|---|---|---|---|---|
| Model monitoringnotice when the world changes and the model quietly gets worse | SageMaker Model Monitor | Azure ML model monitoring | Vertex AI Model Monitoring | Lakehouse Monitoring | Evidently |
Related ideas
Where it shows up
Part of the Run level on these paths: ML Engineering (production ml).