On owned infrastructure, cost was decided at purchase and then largely fixed. Under consumption pricing, cost is a continuous output of engineering decisions, and it drifts upward by default because nothing stops it.
This lands on infrastructure teams who did not sign up for financial management. Here is the minimum that works, without adopting a framework or hiring anyone.
1. Attribution before optimisation
You cannot manage what you cannot attribute. Every resource needs to map to something that has an owner: a team, an application, a cost centre.
Tagging is the mechanism, and tagging fails without enforcement. Policies that reject untagged resources at creation work. Policies that report untagged resources monthly do not, because the report goes to someone who did not create them.
Do this first. Every subsequent practice depends on it, and retrofitting tags onto a sprawling estate is miserable.
2. The waste sweep
Run it monthly. Most of the savings in most organisations are here, and none of it requires architectural change:
- Storage volumes not attached to anything
- Snapshots older than the retention policy nobody enforces
- Instances running that nobody has connected to in weeks
- Non-production environments running at nights and weekends
- Load balancers and addresses left behind by deleted things
- Data in a storage class more expensive than its access pattern justifies
The last one is frequently the largest single item and is invisible unless you look.
3. The floor commitment
Covered in its own post. In short: find the consumption level you never drop below, commit to that at the best rate, leave the uncertain portion uncommitted.
4. Showback
Send each team their own number, monthly, in a form they can act on.
Not a chargeback mechanism, which turns into an accounting argument. Just visibility: here is what your environment cost, here is the trend, here are the three largest items. Teams generally respond to their own number without any enforcement, and they respond much better to it than to a central optimisation initiative.
The anti-pattern
A cost optimisation project. Six weeks of effort, a report, a set of recommendations, savings realised, and then eighteen months of drift until the next one.
Cost management is a routine, not a project. Fifteen minutes a month of waste sweeping beats a quarterly heroic effort, because the drift is continuous and the intervention should be too.
The thing nobody says
Some of the cost is correct and should not be optimised. A system that is expensive because it is doing valuable work at high utilisation is working as intended.
The purpose of this practice is to distinguish spend that produces value from spend that produces nothing, and to eliminate the second. An organisation that has driven its cloud bill down while its systems got slower has not succeeded at anything.
Keep a small number of efficiency measures next to the cost: utilisation, cost per transaction, cost per user. Cost alone is not a metric, it is half of one.