Why your cloud bill grew and traffic did not

Most of it is things nobody is using.

The short answer

Cloud bills usually grow through over-provisioned instances, environments nobody switched off, snapshots and logs with no lifecycle policy, and untracked data transfer. A first audit on an unmanaged account typically finds thirty to fifty percent in savings with no performance impact.

Key takeaways

  • Tag everything by environment and owner, or attribution is guesswork.
  • Right-size from actual utilisation data, not from the size someone picked in year one.
  • Storage and snapshots without lifecycle policies grow forever.
  • Reserve capacity only for workloads whose shape you are confident about.

Cloud spend rarely grows because of a decision. It grows because of the absence of one — nothing was ever switched off, resized or deleted.

Where the money goes

CauseTypical share of wasteFix
Over-provisioned compute30–40%Right-size from utilisation data
Idle or forgotten environments15–25%Schedule shutdown or delete
Unattached storage and snapshots10–20%Lifecycle policies
Logs retained forever5–15%Retention rules and cheaper tiers
Data transfer5–15%CDN, region placement, compression

Tag before you optimise

Without tags for environment, service and owner, cost reports are a single number nobody can act on. Tagging is unglamorous and it is the prerequisite for every other saving on this page.

Reserve carefully

Committed-use discounts are substantial and they are a bet on your architecture staying similar. Right-size first, then reserve the stable baseline, and leave the variable part on demand. Reserving before right-sizing locks in the waste for a year.

Make it a habit

A monthly review with spend attributed by tag, anomalies flagged and one action taken keeps the bill honest. Done once as a project, cloud cost work decays within two quarters — which is why it belongs on the maintenance retainer rather than in a one-off engagement.

Questions people also ask

Done from utilisation data, no. Right-sizing an instance running at eight percent CPU changes nothing a user can perceive. We measure before and after and make the changes incrementally rather than in one sweep.

Keep reading

Related

What this connects to

Rather have this answered about your own account?

Send us what you have. We will look at it properly and write back with what we would change, in the same plain terms as the page you just read.