Design automated cloud cost optimization

WHAT THIS TESTS This evaluates whether you can design FinOps automation that saves real money while treating production safety as the hard constraint, not an afterthought.

A GOOD ANSWER COVERS First classify waste. Idle resources: instances, load balancers, and databases with near-zero utilization. Oversized resources: instances whose CPU and memory headroom far exceeds need, candidates for rightsizing. Orphaned resources: unattached disks, old snapshots, unused IPs, and forgotten dev stacks. Storage lifecycle: cold data sitting in expensive hot tiers that should move to cheaper classes or expire. Commitment gaps: steady baseline usage running on costly on-demand that should be covered by savings plans or reserved capacity. Actions scale with risk: recommend and auto-apply rightsizing where safe, schedule non-production environments to shut down nights and weekends, delete clearly orphaned resources after a grace period, apply storage lifecycle rules, and surface commitment-purchase recommendations. Safeguards are central: drive everything from accurate tagging and ownership so you know what each resource is; scope automatic deletion and shutdown to non-production; require human approval for any production change; run dry-run reports before acting; stage changes with monitoring and automatic rollback; and exempt tagged critical resources.

COMMON WRONG ANSWERS Auto-terminating or downsizing production with no approval or rollback, risking outages. Acting without tags or ownership, so you cannot tell prod from junk. Ignoring commitment-based savings, often the largest lever. Deleting snapshots that are actually backups. No grace period or dry run. Treating cost as more important than availability.

LIKELY FOLLOW-UPS How do tags drive safety? When is recommend-only better than auto-act? How do savings plans differ from reservations? How do you avoid deleting backups by mistake?

ONE CONCRETE EXAMPLE The system schedules all dev and staging instances off at night, auto-deletes volumes unattached for 30 days after notifying owners, rightsizes only non-prod automatically, and for production emits rightsizing and savings-plan recommendations that require an owner's explicit approval before any change is applied, so cost cutting never silently disrupts a live service.