On-premises capacity was fixed, paid for in advance, and visibly finite. Cloud capacity is elastic, billed continuously, and invisible until the invoice. The behavior that produced sensible infrastructure decisions for decades produces runaway spend in the new model, and it does so without anybody making an obviously bad call.
Cost control is therefore mostly an information problem, and the sequence below reflects that. Each step is a prerequisite for the next, and skipping ahead is the usual reason FinOps initiatives produce a report rather than a saving.
One. Attribution before optimization
You cannot reduce a cost nobody owns. Enforce tagging through policy rather than convention, so untagged resources cannot be created, and map every tag to an accountable team.
Expect this step to surface the uncomfortable finding that a material share of spend belongs to nobody. Resolving that is the first saving, and it is usually a large one.
Two. Eliminate the idle
Non-production environments running outside working hours, orphaned storage, unattached volumes, load balancers serving nothing, snapshots nobody will restore. This is unglamorous and it is typically the largest single reduction available.
Automate it rather than running a clean-up campaign, because a campaign is undone within two quarters.
Three. Rightsize against observed use
Most workloads are provisioned for a peak that was estimated rather than measured, and never revisited. Rightsizing needs utilization data over a representative period, including seasonal peaks, which is why it comes after monitoring rather than before.
Four. Commit only what is now predictable
Reservations and savings plans are the most commonly mis-sequenced step. Committing before rightsizing locks in the over-provisioned shape for one to three years, and converts a fixable problem into a contractual one.
Commit to the floor of steady-state demand, not to current consumption.
Five. Make cost visible at design time
Everything above is remediation. The durable change is putting the run-rate consequence of an architectural choice in front of the person making it, in the pull request, before it ships.
That is where cost control stops being a periodic exercise and becomes a property of how the organization builds.
A note on the sustainability overlap
The work above reduces energy consumption for the same reasons it reduces cost, so the two programs are substantially the same program. Where sustainability reporting is an obligation, instrumenting for it alongside cost attribution is far cheaper than doing it separately later.
Measurement methodology varies between providers in scope and basis, so agree the methodology before setting a target.
Discuss a cloud cost program
If this describes something you are working through, we can talk it through against your specific environment.