FinOps: seven levers to cut your cloud bill without slowing delivery
Cloud bills grow from missing visibility and ownership, not from the cloud itself. Seven concrete levers to regain control, ordered from quickest to most structural.
When the cloud bill surprises you every month, the provider is rarely the problem. It's almost always a mix of oversized resources, environments left running idle and, above all, no clear owner for each cost. FinOps is the practice that brings technology, finance and the business together to fix it.
The FinOps Foundation organizes the work into three recurring phases: inform (visibility and allocation), optimize (act on spend) and operate (make it a habit). These are the seven levers with the most impact, from quickest to most structural.
1. Tagging and showback by team and service
Without allocation there's no conversation. Define a minimum tagging standard (service, team, environment, cost center), enforce it in infrastructure as code and publish a cost report per team. When each team sees its own spend, behavior changes.
2. Rightsizing with real data
Review CPU and memory usage at high percentiles (for example p95) over at least two to four weeks. Instances, databases and clusters that never exceed a fraction of their capacity are immediate candidates for downsizing.
3. Turn off what isn't used
Development and test environments don't need to run at night or on weekends. If an environment is used 12 hours a day, five days a week, it's in use 60 of 168 weekly hours: switching it off the rest of the time cuts roughly 64% of the compute cost of those resources.
4. Commitments for the stable baseline
After rightsizing, the load that stays steady is a candidate for commitment models such as Savings Plans, reserved instances or committed-use discounts. Order matters: committing before optimizing means prepaying for waste.
5. Storage lifecycle
Data has a temperature. Set policies that move rarely accessed objects to cheaper tiers, delete orphaned snapshots and volumes, and define clear retention for logs and backups.
6. Transfer-aware architecture
Traffic between zones, regions and out to the internet is usually the least visible cost. Review NAT paths, cross-availability-zone communication and replication patterns. Sometimes an architecture change, such as moving processing closer to the data, saves more than any discount.
7. Unit economics and guardrails
The metric that truly matters isn't the total bill but the cost per business unit: per transaction, per active customer or per processed order. Pair it with budgets and automated alerts, and with infrastructure-as-code policies that block untagged resources or sizes outside the approved list.
A 30-day plan
- Week 1: tagging standard and cost report per team.
- Week 2: rightsize the 20 most expensive resources and schedule non-production environments.
- Week 3: storage cleanup and data transfer review.
- Week 4: define the unit metric, budgets, and decide on commitments.
The goal isn't to spend less at any cost, but to make sure every dollar invested in the cloud has an owner and a return.