Question

You deployed a high-traffic recommendation model on a Vertex AI online endpoint serving roughly 50 million prediction requests per day. You enabled Vertex AI Model Monitoring for feature drift and skew detection. After a month, the finance team flags that logging and monitoring costs have grown far beyond expectations, driven mostly by the volume of request-response logs being captured and analyzed. You must reduce monitoring costs while still reliably detecting statistically significant input drift over daily windows. What is the most appropriate action?

Accepted Answer

Lower the logging sampling rate so only a representative fraction of prediction requests are captured for drift analysis.

Answer

Disable feature drift monitoring entirely and rely only on downstream business KPIs to detect problems.

Answer

Reduce the monitoring frequency to run once per month so fewer analysis jobs execute.

Answer

Switch the endpoint to batch prediction so that request-response logging is no longer billed.