Phase 2: Optimize (Efficiency)

Get more value for every dollar

Once you can see where money is going, you can start reducing waste. The Optimize phase is about finding inefficiencies and eliminating them — but without degrading performance or reliability. The goal is not to spend as little as possible; it is to spend as efficiently as possible.

The FinOps Framework organizes optimization into two distinct levers:

Lever 1: Usage Optimization — Do Less
This means consuming fewer resources, or consuming them more intelligently.

• Terminate idle resources ("Zombie hunting"): Resources that are running but not serving any traffic or workload. A development database that nobody has queried in 45 days. A load balancer with zero targets. These are pure waste.
• Rightsizing: Matching the size of a resource to its actual workload. A VM running at 8% CPU utilization is almost certainly over-provisioned. Downsizing from a c5.4xlarge (16 vCPU) to a c5.xlarge (4 vCPU) can cut that resource's cost by 75% with no performance impact.
• Architectural optimization: Using auto-scaling so you only run capacity when demand exists. Using spot/preemptible instances for fault-tolerant batch workloads. Using serverless (Lambda, Cloud Functions) for intermittent workloads so you pay only per invocation.

Lever 2: Rate Optimization — Pay Less
This means paying a lower price for the same resources you are already using.

• Reserved Instances (RIs): A 1- or 3-year commitment to use a specific instance type in a specific region, in exchange for up to 72% discount versus On-Demand pricing.
• Savings Plans: A more flexible commitment model (AWS-specific) where you commit to a dollar-per-hour spend level (e.g., $10/hr) across any instance type or region, in exchange for up to 66% discount.
• Enterprise Discount Programs (EDP): Negotiated agreements with cloud providers for organizations spending above a threshold (typically $1M+/year), offering additional percentage discounts in exchange for minimum spend commitments.
• Spot / Preemptible Instances: Unused cloud capacity sold at up to 90% discount, but the provider can reclaim the instance with 2 minutes' notice. Suitable for stateless, fault-tolerant workloads like batch processing, CI/CD runners, or ML training jobs.

The critical insight: Usage optimization and rate optimization are not interchangeable. You should always rightsize before committing to Reserved Instances. If you buy a 3-year RI for an oversized instance and then rightsize it, you are locked into paying for capacity you no longer use. The correct order is: Rightsize first, then commit.