Online revenue had grown 60% but cloud spend nearly doubled. Peak events caused cart timeouts while off-peak environments stayed oversized.
Problem
Autoscaling rules fought each other. Marketing spin-up environments were never decommissioned. No owner for tagging.
Approach
We correlated checkout latency with instance metrics, rightsized baseline capacity, and aligned scaling policies to campaign calendar. FinOps review became a standing item in weekly ops.
Result
Peak day completed without severity-1 incidents. Controllable spend down 22% while p95 checkout latency improved. Forecast accuracy for Q4 spend within 8% of actual.
Retail peaks expose provisioning habits that are invisible in average months — FinOps here focused on elasticity policy, commitment coverage, and environment scheduling.
“Peak week ran without emergency scale-ups for the first time in three years. The spend curve finally matched the revenue curve.”
Frequently Asked Questions
Where did the 22% savings come from?
Commitment alignment, non-prod scheduling, rightsizing after peak post-mortem, and retiring duplicate monitoring stacks.
How was peak capacity preserved?
Savings targeted waste layers — not headroom required for trading spikes. Load tests validated peak profiles before changes.
What FinOps routines remained?
Monthly showback, anomaly alerts on daily spend, and steering review of tagging compliance by business unit.
Can smaller retailers benefit?
Principles apply at smaller scale; effort is proportional with phased tagging and quick-win focus first.