Reading a cloud bill properly — a 90-minute audit

Most avoidable cloud spend hides in four line items. You can check all of them in about ninety minutes, and the first pass typically finds between fifteen and thirty per cent.

1. Environments nobody is using

Staging, demo and test environments provisioned for a project that ended, running continuously because nothing ever turned them off. This is the largest and easiest saving in almost every audit we do.

Non-production environments rarely need to exist outside working hours. Scheduling them to stop overnight and at weekends removes roughly two thirds of their cost without anyone noticing.

2. Instances sized for a guess

Almost everything is sized at provisioning, before anyone knows the real load, and almost nothing is revisited. Pull the last month of processor and memory utilisation. Anything sitting under twenty per cent has been paid for twice over.

Resize downward one step and watch for a week. The risk is far lower than teams assume, and the saving is immediate.

3. Storage nothing points at

Snapshots from a migration two years ago. Volumes detached when their instance was deleted. Backups with no retention policy, accumulating for ever.

• List snapshots older than your actual retention requirement.
• List volumes not attached to anything.
• Check whether infrequently accessed data is on storage priced for frequent access.

4. Egress nobody modelled

Data leaving the provider is billed, and it is the line item that surprises people most. Common causes are backups written to another region, chatty inter-service traffic crossing a zone boundary unnecessarily, and media served directly rather than through a content delivery network.

A CDN in front of static assets usually pays for itself immediately, because its egress is cheaper than the origin's and most requests never reach the origin at all.

Making it stick

A single audit is a one-off saving; costs drift back within two quarters. Three habits keep it: tag every resource with an owner so orphans are identifiable, set a budget alert at a threshold that will actually fire, and put a fifteen-minute cost review in an existing recurring meeting rather than creating a new one.