GoldenrozBook a call

Cloud cost optimization checklist: where the money goes and what to cut first

Cloud cost optimization works in a fixed order: first make the bill readable by service, environment and owner, then delete what nobody uses, then shrink what is oversized, then fix storage, logs and data transfer, and only then buy commitments such as savings plans or reservations. Buying commitments first locks in waste. Most of the work is finding out who owns each line, not negotiating price.

A gold sphere under a clear glass dome on a navy base.

Why cloud bills grow

A server you own costs the same whether it works or sits idle. A cloud resource bills for every hour it exists, and creating one takes a minute. Nobody has to approve it, and nobody is reminded to delete it.

So bills grow in a predictable way. A test environment is created for a project and outlives it. An instance is sized for a launch-day peak and never reduced. Logs are kept forever by default. A disk is detached from a deleted server and keeps billing. Each item is small. Together they add up.

The fix is less about price and more about ownership. A line with a named owner gets questioned. A line with no owner does not.

Where the money goes

Most cloud bills fall into six groups.

Group What is in it Typical waste
Compute Virtual machines, containers, serverless functions Oversized instances, non-production running nights and weekends
Databases Managed database instances, replicas, backups Instances sized for peak, unused replicas, long backup retention
Storage Object storage, disks, snapshots Unattached disks, old snapshots, everything in the most expensive tier
Data transfer Traffic leaving the provider, between regions and between zones Chatty services in different zones, traffic through a NAT gateway that could use a private endpoint
Observability Log ingestion, metrics, traces Debug logs in production, unlimited retention, high-cardinality metrics
Licences and support Operating system and database licences, support plans Licences paid twice, a support tier nobody uses

Data transfer and observability surprise people most, because nobody forecasts them. On AWS, for example, a NAT gateway is billed for each hour it runs and for each gigabyte it processes, so a busy private subnet produces a line item that grows with traffic. Log ingestion is billed by volume on all major providers. Check the current rates on each provider’s pricing page for your region.

The checklist, in the order that works

The order matters. Each step changes the baseline for the next.

Step 1: make the bill readable

  • Export twelve months of billing data and group it by service, environment and team.
  • Turn on the provider’s cost tool: AWS Cost Explorer, Microsoft Cost Management or the Google Cloud billing reports.
  • Agree a tagging or labelling standard with at least three tags: owner, environment and application.
  • List the ten largest lines and the ten fastest-growing lines. Write one sentence on what each pays for.
  • Mark every line that nobody can explain. Those come first.

Step 2: delete what nobody uses

  • Disks not attached to any instance
  • Snapshots and machine images older than your retention policy
  • Load balancers with no healthy targets
  • Reserved public IP addresses that are not attached to anything
  • Test and demo environments from finished projects
  • Database replicas nobody reads from
  • Storage buckets holding old build artifacts and exports

Take a final snapshot before deleting anything you are unsure about, and record who approved the deletion. This step carries the least risk and often produces the fastest result.

Step 3: schedule non-production

  • Stop development and staging environments outside working hours where the team’s schedule allows.
  • Scale non-production databases down to the smallest size that works.
  • Give every temporary environment an expiry date at creation.

Simple arithmetic shows why this matters. A week has 168 hours. An environment that runs only 10 hours a day on 5 working days runs 50 hours, which is about 30% of the week. The remaining 70% of its compute hours are paid for and unused. This is an illustration, not a forecast for your account.

Step 4: right-size production

  • Compare each instance’s size with its measured CPU and memory use over at least two weeks, including a busy period.
  • Propose one size down where peak use is well below capacity, and note the risk of each change.
  • Change one system at a time, in a maintenance window, with a way back.
  • Review autoscaling minimums. A minimum set for a launch often stays for years.

Right-sizing a production database deserves care. Read the database’s own metrics, not only the instance’s, and test on staging first.

Step 5: storage, logs and transfer

  • Add lifecycle rules that move old objects to a cheaper tier and delete what is past retention.
  • Set log retention per log group. Thirty days of debug logs is rarely needed.
  • Reduce log volume at the source: no debug level in production.
  • Find traffic that crosses zones or regions without a reason, and traffic that reaches provider services over the public path when a private endpoint exists.

Step 6: buy commitments

  • Measure the steady baseline that remains after steps 2 to 5.
  • Cover part of that baseline with a commitment: Savings Plans or Reserved Instances on AWS, reservations or savings plans on Azure, committed use discounts on Google Cloud.
  • Start with a shorter term and partial coverage if the architecture is likely to change.
  • Record the expiry date and an owner for the renewal decision.

A commitment is a promise to pay whether you use the capacity or not. Bought before cleanup, it locks in the waste you were about to remove. Commitments bought on the wrong baseline are one of the problems we list on our service page.

Step 7: keep it down

  • Set a budget per team or environment, with an alert before it is reached.
  • Turn on anomaly alerts, so an increase is noticed in days.
  • Review the bill by service once a month, with the owners in the room.
  • Reject untagged resources in the pipeline where the platform allows it.

What each cut risks

Action Risk How to lower it
Deleting unattached disks and old snapshots Low Final snapshot, written approval
Stopping non-production at night Low Agree hours with the team, allow a manual start
Storage lifecycle rules Low to medium Check retrieval times and fees of the colder tier
Shortening log retention Medium Confirm audit and compliance retention needs first
Right-sizing production compute Medium One change at a time, measured peak, rollback
Right-sizing a production database Medium to high Test on staging, maintenance window, backup first
Buying a commitment Financial Size to the post-cleanup baseline, partial coverage

FinOps: who owns the bill afterward

A cleanup is a one-time event. Without a habit, the bill returns to its old level within a year. The FinOps Foundation describes the habit as three repeating phases: Inform, Optimize and Operate. In plain terms: see the cost by owner, find the improvements, and make acting on them part of each team’s normal work.

For a mid-size company that means three things. Finance sees the bill by team, not as one number. Each team sees its own line monthly. And someone has the standing job of reading the bill.

If nobody inside the company has that time, the monthly review can sit with an outside provider. Under our managed cloud services retainer, the bill is read by service each month, and idle resources, oversized instances and commitment options are listed for your decision. Patching, backups, access and monitoring are part of the same retainer, which is quoted after an assessment.

Infrastructure as code helps here too, because a resource created through a reviewed pull request has an author and a reason. The workflow is explained in Terraform and CI/CD explained. If part of the estate has steady load on hardware you already own, compare it honestly with staying on-premises, as described in what is hybrid cloud.

What we do, and what we do not claim

Goldenroz does not charge a percentage of savings, does not resell cloud capacity and does not negotiate vendor contracts. We are not a partner of any cloud provider. We do not promise a savings figure before reading the bill, because it would be a guess.

Our cloud cost optimization service is a fixed-price review followed by implementation phases, led by Shai Goldenberg, our CTO. The review produces a bill breakdown, a right-sizing list, a list of idle and orphaned resources with owners, a storage and data transfer review, a commitment plan sized to the post-cleanup baseline, and a tagging and budget standard. It is not a fit for organizations whose annual cloud spend is small next to a $25,000 project.

What to do next

Do steps 1 and 2 yourself this month. They need no outside help and little risk. If the bill is still hard to explain after that, or the remaining changes touch production systems, our cloud cost optimization review gives each change a monthly figure and a risk rating. Projects start at $25,000 at a fixed price per phase, and the first one-hour scoping call is free.

Questions buyers ask

What is cloud cost optimization?

It is the work of finding what a cloud bill pays for, removing what is not needed and buying the remainder at a lower rate. It covers idle resources, oversized instances, storage and log retention, data transfer and commitment discounts, plus the tagging and budgets that keep the bill from growing back.

What should we cut first on a cloud bill?

Start with resources that do nothing: unattached disks, old snapshots, forgotten load balancers and test environments nobody uses. Deleting them carries the least risk. Then schedule non-production environments to stop outside working hours, and only after that change the size of production systems.

Should we buy reserved instances or savings plans?

Yes, for the steady part of your usage, and only after cleanup and right-sizing. A commitment sized to the bill before cleanup pays for capacity you are about to remove. Size it to the baseline that remains, and leave the variable part on pay-as-you-go rates.

How much can cloud cost optimization save?

It depends entirely on the account, and anyone who quotes a percentage before reading your bill is guessing. An account that has never been reviewed usually has idle resources and oversized instances. An account that is already tagged and reviewed monthly has less to find. A review gives each proposed change its own monthly figure.

What is FinOps?

FinOps is the practice of managing cloud spend as a shared responsibility between engineering, finance and the business. The FinOps Foundation describes three phases that repeat: Inform, where you see cost and usage; Optimize, where you find improvements; and Operate, where teams act on them with shared accountability.

Written by

CTO and technology partner

Shai leads the technology side: business software, SaaS and mobile apps, payments and escrow systems, Postgres security, AI agents built with Claude Code and MCP, and IT and infrastructure projects with senior infrastructure engineers.

Keep reading

Talk to the person who will do the work

One hour, no sales team. You talk to Daniel or Shai, we look at your account or your product, and you leave with a clear next step.

Or send the details

Book a 1-hour consultationWhatsApp us