Design Cloud Cost Optimization

Problem Statement: Cloud Cost Optimization Engine

Design a multi-cloud FinOps platform that ingests usage and billing signals (CUR, Cost Management exports, Prometheus/Kubernetes metrics), attributes spend to teams and services, and produces actionable savings—rightsizing, idle shutdown, commitment planning, storage tiering—with optional auto-remediation behind guardrails. Spot, CloudHealth, and Kubecost interviews test whether you treat cost as a data + control-loop problem: observe spend, model waste, recommend changes, verify impact—without breaking production SLOs.

Interview Focus

• Tie every decision to measurable unit economics, attribution accuracy, and safe remediation
• Quantify ingest lag, CUR freshness, and recommendation confidence before automating changes
• Explain human-in-the-loop vs auto-apply and why blast radius caps matter
• Describe failure modes: stale billing data, tag drift, false-positive rightsizing, API throttling

How to open this one

The framing that signals depth on cost optimization is treating it as a data-plus-control-loop problem, not a dashboard: observe spend (billing exports, Kubernetes metrics), attribute it to teams, model waste, recommend changes, and verify impact — with auto-remediation only behind guardrails. Lead with why blast-radius caps and human-in-the-loop matter, and the failure story that proves it: a false-positive rightsizing recommendation auto-applies and throttles a production service. That shows you understand cost engineering must never trade an SLO for a saving.