ModelOps (model operationalization) is the discipline that manages the complete lifecycle of artificial intelligence and machine learning models in production environments, from deployment to retirement. It goes beyond model development: it focuses on governing, monitoring, versioning, and retraining models in a systematic and controlled way once they are operational.
Its importance lies in the fact that a deployed model is not a static asset: its performance degrades over time due to data drift (changes in input data) or concept drift (changes in the relationship the model learned). ModelOps provides the processes to detect and correct these issues, in addition to ensuring traceability, regulatory compliance, and auditing.
In practice, it covers tasks such as:
- Continuous monitoring of accuracy and latency metrics.
- Automated retraining when performance drops.
- Version control and rollback to previous models.
It should be distinguished from MLOps: ModelOps adopts a broader governance perspective, also applicable to analytical or rule-based models, not just machine learning.