Skip to main content
Home
HexagonalGuru

How to Cut Your Cloud Bill by 30% Without Touching Performance

HexagonalGuru's avatar

HexagonalGuru

How to Cut Your Cloud Bill by 30% Without Touching Performance

A project's first AWS, GCP or Azure bill is almost endearing: forty euros. A year later it's four thousand, and the question "why does it go up every month?" is answered with an awkward silence. The honest answer is rarely "because the business is growing": cloud waste accumulates silently, and it bills at full price.

Flexera's annual State of the Cloud reports have put that waste at around 30% of total spend for years. The good news: recovering it doesn't require rewriting anything. It requires method — the FinOps cycle — and five levers that cover most of the savings. It's exactly what we apply in our infrastructure audits, and this guide explains it without the fluff.

The Problem: You Don't Pay for What You Use, You Pay for What You Ask For

The cloud model is on-demand: every resource spins up in seconds and bills until someone deletes it. "Someone" almost never shows up. The pattern we find in nearly every audit:

  • Oversized instances "just in case": servers at 8% CPU, ordered twice as big for a traffic spike that never came back.
  • Dev and staging environments running 24/7, even though the team works 50 hours a week.
  • Orphaned disks from machines deleted months ago, snapshots nobody cleans up, elastic IPs attached to nothing.
  • Storage in the most expensive tier for data nobody has read since last quarter.
  • Egress — outbound traffic — growing without anyone attributing it to a specific service.

None of these lines shows up on any default dashboard. They only show up on the invoice, at the end of the month, already consolidated.

The Core Idea: FinOps Is a Cycle, Not a Project

The most common mistake is treating cloud savings as a one-off "cleanup": a week of effort, some clutter deleted, a small celebration… and six months later the bill is back where it was. FinOps (Cloud Financial Operations) treats cost as just another engineering metric, with a continuous three-phase cycle:

The FinOps cycle: Inform (visibility and tagging), Optimize (rightsizing, scheduling, reservations) and Operate (budgets, alerts and governance), with continuous feedback
  1. Inform: visibility first. Consistent tagging (environment, team, service) and cost allocated per product. Without this, every optimization dies in a "someone uses that, better not touch it".
  2. Optimize: act on the savings levers — the five in the next section.
  3. Operate: budgets with early alerts, a light monthly review and governance: who can create what, at what size, with which mandatory tags.

The output of Operate feeds back into Inform, and the cycle starts over. Each pass is cheaper than the last because the bulk of the waste is already gone.

The Five Levers That Save the Most

Ordered from most immediate to most strategic, with the typical savings we see on the affected resource:

Typical savings per lever: off-hours scheduling ~70%, spot ~70%, storage lifecycle ~50%, reservations ~40%, rightsizing ~30%, egress with CDN ~20%
  1. Switch off what doesn't work at night. A week has 168 hours; a team works about 50. Shutting dev and staging down outside working hours cuts that spend by ~70% in one move, without touching production.
  2. Rightsizing. Measure real usage for two weeks and size the instance to the load, not to the fear. And when migrating, look at ARM families (Graviton on AWS, Ampere elsewhere): same performance, ~20% cheaper.
  3. Usage commitments where the load is stable. Savings Plans, reserved instances or CUDs discount 40-70% in exchange for committing to 1-3 years of usage. Powerful, but only after rightsizing: reserving an oversized instance means freezing the waste.
  4. Storage with a lifecycle. Automatic rules that move objects to cold tiers (S3 Infrequent Access, Glacier) after 30-90 days, plus scheduled cleanup of orphaned snapshots and volumes.
  5. Egress under control. A CDN in front of static content, compression, and an architecture aware of which regions talk to each other. It's the most opaque line on the bill and the least talked about.

A Minimal Example: Dev Switches Itself Off

Lever number one fits in a cron job. Stopping all development instances every day at 8 PM:

# cron: 0 20 * * 1-5 — stop the dev environment every weekday
aws ec2 stop-instances --instance-ids $(
  aws ec2 describe-instances \
    --filters "Name=tag:Environment,Values=dev" \
              "Name=instance-state-name,Values=running" \
    --query "Reservations[].Instances[].InstanceId" \
    --output text
)

Notice the dependency on the Inform phase: the script only works because the instances carry the Environment=dev tag. On Kubernetes the equivalent is a CronJob (or KEDA) scaling the dev namespace deployments to zero — and the morning start-up is another symmetric cron, or simply the first developer who needs it.

What You Gain

  • Immediate, sustainable savings. The first two levers alone usually cut 20-30% off the bill in the first month — and the cycle keeps it from coming back.
  • Cost per product, not per provider. Knowing what each service costs turns "the AWS bill" into business decisions: per-customer margins, pricing, priorities.
  • Budgets that hold. With early alerts, a spike on day 3 is detected on day 3, not day 30.
  • Culture. Cost stops being "management's thing" and becomes a metric the team sees, like latency or error rates.

When It's Not the Answer

  • Small bill. If you spend less than ~€1,000/month, engineering hours cost more than the possible savings. Apply the two nearly-free levers (night-time shutdown + budget alerts) and spend the time on product.
  • Hypergrowth. If you're scaling a launch, the opportunity cost of distracting the team outweighs the savings. Optimizing prematurely is just another form of waste.
  • The problem is architecture, not tuning. An application that can't scale horizontally won't be fixed by reservations or rightsizing: savings there come from redesigning, not from switching things off.

Common Mistakes

  1. Buying reservations before rightsizing. The most expensive mistake: you commit to 1-3 years of paying for capacity you don't need.
  2. Optimizing once and never looking again. Without the Operate cycle, waste returns within months: new services, new environments, same habits.
  3. Starting without tags. No cost allocation means no owners, and without owners every deletion proposal dies in a meeting.
  4. Only watching compute. In many workloads, storage and egress are a third to half of the bill. Optimizing only the machines is sweeping half the floor.
  5. Spot in critical production without a plan. Interruptible instances save up to 70%, but only for workloads that tolerate interruptions: batch, CI, async processing.
  6. Reviewing the bill at month-end. By then the damage is done and consolidated. Budget alerts at 50% and 80% are the cheap version of sleeping well.

In Summary

Cutting a cloud bill by 30% isn't a heroic project: it's visibility (tags and cost per service), five levers applied in the right order — night-time shutdown, rightsizing, usage commitments, storage lifecycle and egress — and a light FinOps cycle that keeps the waste from coming back. The cloud isn't expensive; not looking is.

If your bill grows every month and nobody knows exactly why, that's exactly the kind of audit we do: we measure, tag, apply the levers and leave the cycle running for you. Let's talk about your infrastructure →

Keep reading: How much does custom software cost? An honest guide to budgets.

  • #devops
  • #cloud
  • #finops

Related articles

Shall we start?

Ready to build something that grows with your business?

Tell us your goals and together we will map the design, development and technology route that takes you from vision to measurable results.