Glossary

What is Data Drift?

Data drift is when the data feeding a production AI system changes over time, so the model’s inputs no longer match what it was built and validated on — even though the model code is unchanged. The shift can hit feature distributions, schemas, value ranges, or upstream pipelines, and it quietly degrades accuracy until someone notices the downstream impact.

Drift is hard to act on when the data state behind each run isn’t fixed. If you can compare a live dataset against a released, AI-ready baseline, you can see exactly which fields and distributions moved and reproduce the earlier state to confirm the cause.

Frequently asked questions

What causes data drift?

Changes in upstream data sources, pipeline updates, schema edits, shifting user behavior, or new data windows. The model itself stays the same while its inputs move.

How do you detect data drift?

By comparing current production data against a fixed, released baseline state and surfacing which fields, distributions, or schemas have changed.

What is the difference between data drift and concept drift?

Data drift is a change in the input data's distribution; concept drift is a change in the relationship between inputs and the target the model predicts.