Chapter 6. Monitoring ML Applications
According to The Institute for Ethical AI & Machine Learning, 42% of ML projects do not have monitoring in place (see Figure 6-1). Even though that figure is better than the one from the year before (50% with no monitoring), it shows how immature most companies are when it comes to MLOps.
Figure 6-1. Updated image from the State of Production ML in 2025 survey
But monitoring isn’t a “nice-to-have.” Without it, models fail silently. Predictions degrade. Bias can increase over time. Infrastructure costs can spiral. And perhaps most damaging of all, stakeholders will lose trust in the system when performance suddenly drops and no one knows why. In regulated industries, lack of monitoring can even create legal and compliance risk.
By the end of this chapter, you will understand all of the following:
-
The different layers of ML monitoring
-
How to design a practical monitoring strategy
-
What metrics actually matter in production
-
How to set alerts on aspects that matter
-
How monitoring connects to retraining and continuous improvement
In short, this chapter will move you from “Model deployed” to “Model managed.”
What to Monitor
Monitoring requirements for ML projects are highly use-case dependent. While generic software metrics are important, ML-specific monitoring focuses on the model’s relationship with changing data and its ultimate ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access