Cloud cost optimisation gets discussed as if it were a sophisticated discipline. In practice, the first 40% of savings on most accounts comes from four unglamorous things nobody got around to.
Here they are, roughly in order of what they return.
1. Instances provisioned for a load that never arrived
The most common finding, and usually the largest. Someone sized the environment for projected traffic, that projection was optimistic, and nothing was ever revisited.
Look at CPU and memory utilisation over 30 days. If average utilisation sits below 20% and peak below 50%, the instance is at least one size too large — frequently two. Halving an instance size halves its cost, and the change is reversible in minutes.
Do this per workload rather than globally. Databases and batch processing have different profiles from web servers, and a single blanket policy will either under-serve one or over-serve the other.
2. Storage nobody deleted
Storage waste accumulates silently because no single item is expensive.
- Orphaned volumes left behind when instances were terminated. They keep billing indefinitely.
- Snapshots without a retention policy. We routinely find years of daily snapshots where the recovery requirement is 30 days.
- Object storage in the wrong tier. Data accessed twice a year sitting in standard storage instead of an archival class.
- Old machine images from deployments long superseded.
- Load balancer and application logs with no expiry rule, growing forever.
None of these are difficult. They just require someone to look.
3. Static capacity where demand is variable
If your traffic has a daily or weekly shape — and almost all business traffic does — running fixed capacity means paying peak rates around the clock.
Autoscaling groups with sensible thresholds address this. So does scheduled scaling for predictable patterns: development and staging environments rarely need to run at 3am on a Sunday, and shutting them outside working hours removes roughly two-thirds of their cost.
Combine this with commitment discounts for the genuinely steady baseline. Reserved instances or savings plans on the floor of your usage, on-demand for the peaks above it. Committing to the peak is a common and expensive error.
4. Resources nobody can attribute
Untagged resources are not directly a cost problem — they are a diagnosis problem. If you cannot tell which team, environment or client a resource belongs to, you cannot decide whether it is still needed, and the safe default is always to leave it running.
Enforce a minimal tagging policy — owner, environment, project — at creation time. Then a monthly review becomes a filter rather than an investigation.
The things people reach for first, that matter less
Switching cloud provider. Migration cost usually exceeds the price differential, and the waste travels with you.
Micro-optimising compute pricing while ignoring data transfer. Egress charges are frequently a larger line than the instances themselves, particularly for media-heavy applications. A CDN in front of your origin is often the single highest-return change available.
Serverless everything. Excellent for spiky, event-driven work. Meaningfully more expensive than a right-sized instance for steady, predictable load.
Making it stick
One-off optimisation decays. Within six months the same patterns reappear because nothing changed structurally.
What actually holds:
- Budget alerts at meaningful thresholds, routed to someone who owns the response.
- A monthly cost review as a standing item, not an annual panic.
- Tagging enforced at provisioning, not retrofitted.
- Non-production environments scheduled off by default.
- Someone named as accountable for the number.
That last one matters most. Cloud spend that is everybody's concern is nobody's job.
What this looks like in practice
On a recent logistics engagement the combination of right-sizing, autoscaling and storage cleanup reduced monthly infrastructure cost by 41% — while simultaneously removing a single point of failure that had already cost the client a full trading day.
Reducing cost and improving resilience are not usually opposed. Over-provisioned single-region infrastructure tends to be both expensive and fragile.
Our cloud services team runs this assessment as a standalone engagement, and it typically pays for itself in the first month.
Need help with this?
Our team applies exactly this analysis for clients across six countries. Start with a no-cost assessment of your own site.
Request an Assessment