The alert is initiated at 15% free. The clean-up happens at ~10%.
You can improve your response by create an alert that is -recurring- for the Administrative Alert > System subtypes. This recurring alert won't let you forget.
Also, you could add a KPI_HT for the disk space on the UI/Analytics volumes to exclusively alert you when <15% free... this can be done at the VM level or leverage the vC Ops Application container and alert from there.