Not all unit metrics are created equal. They exist on a spectrum from raw infrastructure metrics to pure business metrics. Choosing the right level depends on your audience and the decision you are trying to inform.
Level 1: Resource Metrics (Least Valuable for Business Decisions)
- Examples: Cost per vCPU-hour, Cost per GB of Storage, Cost per Network Egress GB.
- Audience: Infrastructure procurement teams.
- Why limited: These tell you whether you are buying cloud resources cheaply (e.g., using Spot Instances vs. On-Demand), but they say nothing about whether your application is efficient. You could have a very low "Cost per vCPU-hour" while your application wastes 90% of those vCPUs sitting idle.
Level 2: Engineering Proxy Metrics (Good)
- Examples: Cost per API Call, Cost per CI/CD Pipeline Run, Cost per Search Query, Cost per Database Transaction.
- Audience: Engineering teams and Platform teams.
- Why useful: These bridge the gap between infrastructure and business. Engineers understand them intuitively. If you optimize a slow SQL query, the "Cost per Transaction" goes down. These metrics make optimization work visible and measurable.
Level 3: Business Metrics (Best for Executive Reporting)
- Examples: Cost per Customer, Cost per Ride, Cost per Delivery, Cost per Subscriber.
- Audience: CFO, Board of Directors, business unit owners.
- Why powerful: These speak the language of the P&L statement. They allow the business to calculate gross margins directly. A CFO can immediately understand "our cost to serve each customer dropped from $4.50 to $3.80 this quarter" — they cannot interpret "our average EC2 utilization improved by 12%."
Choosing Your Level in Practice
In practice, you need metrics at multiple levels simultaneously. A Platform Engineer optimizing a Kubernetes cluster works with Level 1 and Level 2 metrics daily. The same engineer presenting quarterly results to the VP of Engineering uses Level 2 and Level 3 metrics. The CFO presentation uses only Level 3.
Interviewers often ask: "What FinOps metrics would you track for a SaaS platform?" A weak answer lists AWS Cost Explorer categories. A strong answer describes a layered approach: Level 1 metrics for infrastructure efficiency (utilization rates, Spot vs. On-Demand ratio), Level 2 metrics for engineering optimization (cost per API call by service), and Level 3 metrics for business reporting (cost per tenant, cost per active seat). Explain that each layer serves a different audience and decision type.