Cloud Cost Reduction: The 8 Optimisations That Actually Move the Needle

The Flexera State of Cloud 2026 report finds that organisations waste an average of 32% of their cloud spend. For a company paying $500K/year in AWS or Azure costs, that is $160,000 in annual waste that is not buying performance, reliability, or competitive advantage — it is just padding someone's cloud provider revenue.

The good news: most cloud waste is recoverable without architectural rewrites. The bad news: it requires someone to actually look at the bill and understand what they are seeing.

This guide covers the eight optimisations that reliably deliver the largest reduction in cloud spend, with real benchmarks from production environments.

Why Cloud Bills Spiral

Cloud costs grow because of three failure modes:

1. Optimistic provisioning: Engineers provision compute for peak load (the "it might spike" conversation) and never revisit once the system stabilises at 20% utilisation.

2. Orphaned resources: Dev environments, test databases, EBS volumes, and load balancers created for projects that ended — and never deleted.

3. No ownership model: Nobody has a financial incentive to reduce cloud costs because teams aren't allocated their spend. The engineering team builds, the finance team pays, and nobody connects the two until the CFO asks why cloud costs tripled.

Optimisation 1: Right-Size Over-Provisioned Instances (Saves 20–35%)

AWS Compute Optimiser and equivalent tools on GCP and Azure analyse actual CPU and memory utilisation over 14 days and recommend downsizing.

The benchmark: the average production workload runs at 15–25% average CPU utilisation. Most instances were provisioned for peak, not average. Moving from m5.4xlarge to m5.2xlarge on a set of application servers running at 18% CPU saves 50% of the instance cost, with no performance impact.

How to do it:

1. Enable AWS Compute Optimiser (free)
2. Export recommendations as CSV
3. Test downsize in staging with load simulation
4. Roll out to production instances in off-peak window

Caution: Don't right-size databases without careful testing. CPU utilisation is not the right metric for RDS — look at I/O saturation and connection counts.

Optimisation 2: Reserved Instances and Savings Plans (Saves 30–72%)

On-demand pricing is AWS's most expensive pricing model. For workloads running 24/7 (production databases, core application servers), Reserved Instances (1-year or 3-year commitment) or Compute Savings Plans deliver:

• 1-year No Upfront RI: ~40% savings vs. on-demand
• 1-year All Upfront RI: ~40–45% savings
• 3-year All Upfront RI: ~60–72% savings

The strategy: Cover your baseline compute (the instances you know will run continuously) with Savings Plans. Use on-demand for variable and unpredictable workloads.

Common mistake: Over-committing on Reserved Instances for infrastructure that will change. Savings Plans are more flexible — they apply to any EC2 instance family in any region.

Optimisation 3: Spot/Preemptible Instances for Fault-Tolerant Workloads (Saves 60–90%)

AWS Spot Instances are spare EC2 capacity sold at 60–90% discount. They can be reclaimed with 2-minute notice when AWS needs the capacity back.

This makes them unsuitable for stateful production workloads but excellent for:

• CI/CD build runners
• ML training jobs
• Batch processing pipelines
• Non-critical background workers

Implementation: Configure a Spot Instance Request with an On-Demand fallback. AWS Auto Scaling Groups support mixed instance policies — e.g., 70% Spot, 30% On-Demand — providing cost savings without a single point of failure.

A data pipeline running on Spot Instances costs $0.018/hr vs. $0.192/hr on-demand for a c5.2xlarge — a 90% saving.

Optimisation 4: S3 Storage Lifecycle Policies (Saves 40–70% on Storage)

S3 Standard costs $0.023/GB/month. S3 Intelligent-Tiering, Glacier Instant Retrieval, and S3 Glacier Deep Archive cost $0.004–$0.0036/GB/month for infrequently accessed data.

Most organisations pay S3 Standard rates for:

• Log files never accessed after 7 days
• Backup archives that haven't been touched in 2 years
• Build artefacts from 3 release cycles ago

The fix: Implement S3 Lifecycle Policies:

Transition to S3-IA after 30 days
Transition to Glacier Instant Retrieval after 90 days
Expire (delete) after 365 days (for logs/temp data)

For 100TB of S3 storage, moving 60% to Glacier tiers saves approximately $1,200/month.

Optimisation 5: Shut Down Non-Production Environments (Saves 50–70%)

Development, staging, and QA environments typically run 24/7 but are only used 8–10 hours per weekday. That means they run at zero utilisation 65–70% of the time.

Implementation: Lambda or CloudWatch Events-scheduled stop/start scripts:

• Stop all dev/staging EC2 and RDS instances at 8 PM local business hours
• Start at 7 AM Mon–Fri
• Tag-based policy: any instance tagged env=staging follows the schedule

Savings: A $15,000/month staging environment operating 50 hours/week instead of 168 hours/week costs $4,400/month — a $10,600/month saving.

Optimisation 6: Delete Orphaned Resources

AWS Cost Explorer and cloud cost tools like Infracost or CloudHealth identify orphaned:

• EBS volumes: Volumes not attached to any instance, still billed at $0.10/GB/month
• Elastic IPs: Unattached IPs billed at $0.005/hr
• Idle Load Balancers: ALBs with no targets cost ~$16/month each
• Unused NAT Gateways: Each gateway costs ~$45/month plus data processing charges
• Forgotten RDS snapshots: Accumulate at $0.095/GB/month indefinitely

Run a quarterly orphan audit. It typically yields $500–$5,000/month in savings for mid-size AWS accounts with no effort beyond deletion.

Optimisation 7: Multi-Region Architecture Review

Multi-region deployments are appropriate for applications with strict latency requirements or regulatory data residency obligations. For many SaaS products, the "multi-region active-active" architecture was implemented for resilience — but a single-region active + multi-AZ passive configuration delivers 99.99% availability at significantly lower cost.

Data transfer costs are the hidden multiplier: inter-region data transfer costs $0.02/GB vs. $0.01/GB inter-AZ vs. $0.00/GB intra-AZ. A workload pushing 10TB/day cross-region accumulates $6,000/month in data transfer alone.

Review: Does your SLA actually require active-active multi-region? For most B2B SaaS products, the honest answer is no.

Optimisation 8: FinOps Governance Model

The above seven optimisations are one-time wins. Without governance, cloud costs drift back up within 6–12 months as new infrastructure is provisioned without cost awareness.

A FinOps governance model includes:

• Cost allocation tags: Every resource tagged with team, env, product, owner
• Team-level budgets: Each engineering team has a monthly budget and receives alerts at 80% consumption
• Architecture review gate: Any new infrastructure that adds >$500/month requires a brief cost justification
• Monthly FinOps review: 30-minute engineering/finance sync to review top cost movers

The FinOps Foundation's practitioner framework calls this the "Crawl, Walk, Run" maturity model. Most organisations at the "Crawl" stage (reactive cost review) can reach "Walk" (proactive optimisation) within a quarter.

Realistic Savings by Company Stage

Monthly Cloud Spend	Typical Waste %	Recoverable in 90 Days
$10K–$50K	25–35%	$2,500–$17,500/mo
$50K–$200K	30–40%	$15,000–$80,000/mo
$200K–$1M	28–38%	$56,000–$380,000/mo
$1M+	25–35%	$250,000–$350,000+/mo

When to Engage a Cloud Cost Consultant

Self-service FinOps works well for engineering-driven organisations with clear ownership. Cloud cost consulting delivers faster, larger savings when:

• Your team doesn't have dedicated DevOps/FinOps bandwidth
• Your architecture has significant technical debt in cloud provisioning patterns
• You are preparing for a fundraise and need to demonstrate operational efficiency
• Your AWS bill has grown >50% year-over-year without equivalent revenue growth

Ortem Technologies' Cloud & DevOps practice offers cloud cost assessment engagements: a 2-week audit that identifies every recoverable dollar, a prioritised remediation roadmap, and implementation support.

Request a Cloud Cost Assessment | Explore Our Cloud & DevOps Services