Rate Optimization: Commitments and Spot

Rate optimisation lowers the price per unit: cover the steady baseline with Savings Plans or reservations sized by coverage and utilisation targets, check break-even, use spot for interruptible work and negotiate enterprise discounts.

After removing waste, the remaining steady usage can be bought cheaper per unit. Rate optimisation is usually run centrally by the FinOps team with procurement and finance, because commitments are shared financial decisions.

flowchart LR
  USE["Usage after clean-up"] --> BASE["Find the steady baseline"]
  BASE --> COMMIT["Cover baseline: Savings Plans / reservations"]
  USE --> VAR["Variable peaks: on-demand"]
  USE --> INT["Interruptible: spot"]
  COMMIT --> KPI["Track coverage and utilisation"]
  KPI --> BASE
  %% step USE: Start from usage after rightsizing, or you commit to waste.
  %% step BASE: The baseline is the level usage almost never drops below, hour by hour.
  %% step COMMIT: Commit only to the baseline; buy in smaller steps over time (laddering).
  %% step VAR: Peaks above the baseline stay on-demand or autoscale.
  %% step INT: Batch, CI and stateless workers can use spot for the deepest discounts.
  %% step KPI: Coverage shows how much is discounted; utilisation shows how much commitment is wasted.

Commitment types (examples)

Provider	Flexible	Specific
AWS	Compute Savings Plans (EC2, Fargate, Lambda)	EC2 Instance Savings Plans, Reserved Instances (RDS, ElastiCache, OpenSearch, Redshift...)
Azure	Savings plan for compute	Reservations (VMs, SQL, Cosmos DB...)
Google Cloud	Spend-based committed use discounts	Resource-based CUDs; sustained use discounts apply automatically

Flexible commitments discount less but survive architecture changes; specific ones discount more but can be stranded.

Two KPIs, one balance

Coverage = discounted eligible usage ÷ eligible usage. Low coverage → paying on-demand for steady load.
Utilisation = commitment used ÷ commitment bought. Below 100% → paying for nothing. Aim high on both; pushing coverage to 100% usually drops utilisation.

Break-even

A 1-year no-upfront commitment at 30% discount pays off if the usage runs more than about 70% of the hours it covers (1 − 0.30). If a workload may disappear in six months, the flexible or shorter option is safer.

Spot

Up to ~90% off, but capacity can be reclaimed with short notice (two minutes on AWS). Use it for fault-tolerant work, spread across several instance types and zones, and handle interruption signals.

Other levers

Enterprise discount programmes and private pricing, marketplace purchases that count towards commitments, and licence choices (bring-your-own-licence, Linux vs Windows, open-source databases).

Traps: buying three-year commitments for workloads planned to migrate; letting teams buy commitments independently (overlap and waste); and comparing savings using billed cost in the month of an upfront payment.