What we look for in a cloud bill

Before you open the console

Get these two things first or the rest is guesswork.

• Cost and Usage Report / BigQuery billing export enabled

  Not the console graphs. The line-item export is the only place the real answers live.

• A read role in the accounts, not just billing

  You cannot tell an idle NAT gateway from a busy one out of billing data alone.

The three that pay for the exercise

In our experience these three cover most of a first pass.

• NAT gateways, one per availability zone

  ₹3,100 a month each before a byte crosses. Three zones is ₹9,400 a month idle, ₹1.1 lakh a year. Ask whether all three are load-bearing.

• gp2 volumes never moved to gp3

  Around 20% cheaper for the same disk, 3,000 IOPS baseline, no downtime, one command.

• Cross-AZ replication nobody designed

  500 GB a day between zones is roughly ₹28,600 a month (₹3.4 lakh a year), usually because a subnet was picked in a hurry.

Data transfer: the section nobody opens

Above roughly ₹1 lakh a month of spend this is usually 10–30% of the bill.

• NAT data processing separated from NAT hours

  The per-GB charge is usually the larger of the two.

• Internet egress by service

  Roughly ₹8.6/GB. Combined with NAT processing it is about ₹12.9/GB.

• Cross-region transfer

  Often a DR copy nobody has reviewed since it was set up.

• Traffic that could use a VPC endpoint

  S3 and DynamoDB traffic routed through NAT is pure waste.

• Public IPv4 addresses

  Chargeable since Feb 2024. Count the unattached ones.

Compute

• Instances under 10% average CPU for 30 days

  Rightsize before you commit to anything.

• Instances running outside business hours that needn't be

  Dev and staging first.

• Previous-generation instance families

  Newer families are usually cheaper per unit of work.

• Orphaned load balancers

  An ALB with no healthy targets still bills.

• Over-provisioned managed databases

  Look at connections and IOPS, not just CPU.

Storage

• Snapshot policies with no expiry

  A slow leak that compounds every month.

• EBS volumes not attached to anything

  Often left behind when an instance is terminated.

• Object storage with no lifecycle rule

  Infrequent-access and archive tiers exist for a reason.

• Duplicate backups across two tools

  Common after a migration nobody finished.

• Log retention set to 'never expire'

  Ingest is one cost; keeping it forever is another.

Commitments: last, not first

Buying these before rightsizing locks in the waste for the full term.

• Current Savings Plan / CUD coverage and utilisation

  Utilisation below ~95% means you are paying for capacity you are not using.

• Anything expiring in the next 90 days

  The renewal is your chance to resize the commitment.

• Commitment term against your actual planning horizon

  A three-year commitment on a workload you may retire in eighteen months is a trap.

Governance

Not savings today, but the reason the savings come back.

• Tagging that lets you attribute cost to a team

  Untagged spend cannot be governed.

• A budget alarm that a person actually receives

  Not one that emails a shared mailbox nobody opens.

• Anomaly detection on spend

  So a surprise is a Tuesday alert rather than a month-end shock.