FinOps and Cloud Cost Optimization

The FinOps lifecycle (inform, optimize, operate), unit economics, tagging, commitment discounts, rightsizing, architecture-level cost levers and showback.

What it is

FinOps (as defined by the FinOps Foundation) is an operating model that brings financial accountability to variable cloud spend through collaboration between engineering, finance and business. Its lifecycle phases are Inform → Optimize → Operate.

Inform: visibility

  • Mandatory tags/labels: application, team, environment, cost center.
  • Cost allocation including shared costs (networking, platform) by usage-based keys.
  • Showback/chargeback dashboards per team.
  • Unit economics: cost per order, per active user, per GB processed, per model inference.

Optimize: levers by impact

Lever Typical saving Notes
Turn off idle and non-prod outside hours 20–70% of non-prod Scheduling, scale-to-zero
Rightsizing 10–30% From utilization data; ARM instances (Graviton/Axion/Cobalt)
Commitment discounts 20–70% vs on-demand Savings Plans/Reserved Instances, Azure reservations/savings plans, GCP committed use discounts; cover the steady baseline only
Spot/preemptible Up to ~90% Stateless, fault-tolerant, batch workloads
Storage tiering and lifecycle Large for data-heavy apps Archive tiers, delete old snapshots/logs
Data transfer Often hidden Keep traffic in-AZ/region, use CDNs, VPC endpoints instead of NAT
Architecture changes Largest long-term Serverless for spiky loads, caching, efficient data formats

Operate: continuous practice

Budgets and anomaly alerts, cost reviews in sprint/quarterly rituals, cost as an architecture review criterion, KPIs per team.

flowchart LR
  I[Inform<br/>allocation, unit cost] --> O[Optimize<br/>rightsizing, commitments,<br/>architecture]
  O --> P[Operate<br/>budgets, reviews, policy]
  P --> I

[!CTO] Report unit cost trends (cost per transaction) rather than total spend; growing spend is fine if unit cost falls and revenue grows.

Interview questions and answers

The cloud bill rose 40% this quarter. What do you do?

Break it down by service, account and tag to find the drivers; separate growth-driven increases (more customers) from waste (idle resources, unbounded logs, data transfer, misconfigured autoscaling); fix quick wins; set budgets and anomaly alerts; address architectural causes; report unit cost.

How much commitment should you buy?

Cover the stable baseline seen over recent months (often 60–80% of steady usage), staggered terms, and review monthly; leave peaks to on-demand and Spot.