Cloud cost optimisation for Swiss SMEs: FinOps basics

This is the practitioner’s version of cloud cost work — what to do, in what order, with your hands on the console. The leadership framing sits in a CEO’s guide to cloud cost optimisation.

Where the waste actually is

Before optimising, it helps to know that cloud waste is boringly predictable. In most SME estates it concentrates in five places:

• Non-production environments running around the clock — used roughly 45 hours a week, billed for 168
• Instances sized from the old server, which was itself sized for a peak that no longer occurs
• Orphaned storage — unattached volumes, snapshots from migrations that finished, backups of systems that were retired
• Data transfer, especially between availability zones in architectures that were never designed with transfer cost in mind
• Managed services chosen for convenience where the self-operated equivalent was adequate

None of it is exotic, and all of it accumulates silently because nothing forces a review.

FinOps practices, ordered by payback

Make the bill legible first

Tag every resource with a team, product or cost centre, and produce a monthly breakdown. This is not itself a saving and it is still the highest-return work available, for a behavioural reason: teams that see their own number change what they provision, without anyone instructing them to.

Set a coverage target — around 80% of spend attributable — and treat untagged resources as a defect to be fixed, not a report to be filtered.

Schedule the environments that should not run

Development, test and staging environments outside working hours are the largest single reversible saving in most SME estates. Implementation is a scheduled start/stop, and the risk is close to zero because the environments are recreatable by definition.

Start here if you want a result inside a week.

Right-size against measured usage

Collect at least a month of CPU, memory and I/O data before resizing anything. Sizing from intuition reproduces the original error in the other direction, and an under-sized production workload costs far more than it saves.

Resize in production last, after non-production has demonstrated the new sizing holds.

Clean up storage on a schedule

Orphaned volumes and old snapshots do not announce themselves. A quarterly sweep with a simple rule — anything unattached for 30 days, anything older than the retention policy — removes a line item that otherwise only grows.

Careful with one exception: retention obligations. Under revDSG/nDSG and GDPR some data has a minimum retention and some has a maximum, and a cleanup script that ignores both creates a compliance problem while saving a modest amount of money.

Commit only to what is genuinely steady

Reserved capacity and savings plans reduce the rate for predictable workloads. Do this last, after right-sizing, or you will commit to capacity you then optimise away and pay for both.

A reasonable rule for an SME: commit to the baseline you are confident persists for the full term, and leave the variable portion on demand.

Budgeting a variable bill

The structural difficulty for an SME is not the size of the cloud bill but its variability — it replaces a known depreciation schedule with something that moves monthly.

What makes it manageable:

• Budget alerts at thresholds, not at the limit. An alert at 100% of budget is a notification of something already happened.
• A forecast that separates baseline from variable. The baseline is predictable; the variable part should be explainable by a business driver. If it is not, that is the finding.
• Unit cost as the reported metric. Cost per customer or per transaction, not total. A growing business should spend more; unit cost tells you whether it is spending well.
• A monthly variance conversation with a named owner. Fifteen minutes, with the question “what changed and why”.

What not to do

• Do not cut logging or monitoring to save money. It looks efficient and removes the evidence you need during an incident — and for regulated workloads, part of what you are obliged to retain.
• Do not buy a cost tool before doing the tagging. The tool will show you an unattributable bill in a nicer interface.
• Do not run a one-off cleanup. Costs drift back within two quarters unless something recurring resists it.
• Do not optimise a workload that is about to move. Tune it after it lands.

What realistic savings look like

Honest framing rather than a headline number: the savings available depend entirely on how long the estate has run without cost governance. An estate that has never been reviewed usually has a substantial amount of the five waste categories above; one that is reviewed quarterly has little.

The useful expectation is that the first review produces a step change and subsequent ones produce maintenance. If you are promised a recurring large percentage every year, the promise is about the first year repeated.

Related reading

• Swiss cloud adoption: five lessons on getting to real ROI
• Cloud migration roadmap: a decision framework
• Cloud & DevOps

If you want a review of where your spend is actually going, book a call. No obligation.