Generic Advice Won’t Cut Your Bill
Every AWS cost optimization article says the same things: use reserved instances, rightsize your compute, delete unused resources. That advice isn’t wrong — it’s just not specific enough to act on.
Here’s what actually works in 2026, with specific tools and tactics for each category.
1. Rightsizing: Stop Guessing, Start Measuring
The problem: Teams pick instance types based on initial load testing or gut feel, then never revisit the decision.
The fix: Use 14-day CPU and memory utilization data to rightsize. The threshold that matters:
- Below 20% average CPU → downsize by one instance class
- Below 5% average CPU → evaluate whether the workload should exist at all
- Memory utilization below 30% → switch to a compute-optimized instance instead of general purpose
The tool: AWS Compute Optimizer gives free recommendations. IAN goes further by correlating utilization with deployment patterns — catching resources that spike during deploys but are idle 95% of the time.
Typical savings: 20-35% of compute costs.
2. Reserved Instances and Savings Plans: The Math
The problem: Teams avoid commitments because “what if our usage changes?” Meanwhile, they’re paying on-demand prices for workloads that have run steadily for a year.
The rules:
- If a workload has run for 3+ months at stable utilization → Compute Savings Plan (flexible, 20-30% savings)
- If a workload has run for 6+ months at stable utilization → EC2 Instance Savings Plan (less flexible, 30-40% savings)
- If a workload has run for 12+ months with no changes planned → Reserved Instances (least flexible, 40-60% savings)
Start with Compute Savings Plans — they cover EC2, Fargate, and Lambda, and you can change instance types freely.
Typical savings: 25-45% on committed workloads.
3. Spot Instances: Not Just for Batch Jobs
The problem: Teams think Spot is only for batch processing. In reality, Spot works for any workload that can handle interruptions with grace.
Good Spot candidates:
- CI/CD runners (save 60-80% on build compute)
- Development and staging environments
- Stateless web servers behind an ALB (with on-demand fallback)
- Data processing and ETL
The tactic: Use a diversified Spot strategy across multiple instance types and availability zones. Set up Spot Fleet or Karpenter (for Kubernetes) to automatically fall back to on-demand when Spot capacity is unavailable.
Typical savings: 60-80% on eligible workloads.
See the IAN team run on your cloud. We connect to your AWS account via a scoped read-only role, run the Observe-tier agents, and leave you with a concrete audit report — cost waste, security exposure, compliance gaps, and a labor-offset estimate. You keep the findings regardless of next steps. Get a free infrastructure audit →
4. Storage Lifecycle Policies: Set and Forget
The problem: S3 buckets accumulate data forever. Nobody sets lifecycle policies because they’re afraid of deleting something important.
The rules:
- Logs older than 30 days → S3 Infrequent Access (45% cheaper)
- Logs older than 90 days → S3 Glacier Instant Retrieval (68% cheaper)
- Logs older than 1 year → S3 Glacier Deep Archive (95% cheaper)
- EBS snapshots older than 30 days for non-production → delete
- Unattached EBS volumes → snapshot and delete
Typical savings: 40-70% on storage costs.
5. Network Costs: The Hidden Line Item
The problem: Data transfer costs sneak up on teams. Cross-AZ traffic, NAT Gateway charges, and CloudFront invalidations add up.
Quick wins:
- Use VPC endpoints for S3 and DynamoDB (eliminates NAT Gateway charges for those services)
- Consolidate services in the same AZ where possible (cross-AZ costs $0.01/GB each way)
- Use S3 Transfer Acceleration only when you actually need it (it costs 4x normal transfer)
- Review NAT Gateway throughput — a single busy NAT Gateway can cost $1,000+/mo
Typical savings: 10-20% of network costs.
6. AI-Powered Continuous Optimization
The problem with manual optimization is that it’s a point-in-time exercise. You rightsize today, and three months later usage patterns have changed and you’re overprovisioned again.
AI-powered cost monitoring solves this by:
- Scanning continuously — catching waste as it appears, not months later
- Correlating with deployments — understanding which changes caused cost increases
- Generating specific actions — not just “you’re overspending” but “downsize i-0abc123 from m5.2xlarge to m5.xlarge, estimated savings $340/mo”
- Tracking savings over time — proving the ROI of optimization work
IAN monitors your AWS accounts continuously and generates a prioritized list of savings opportunities with specific actions. Most teams find enough waste in the first audit to pay for the platform for a full year.
Start Optimizing
Connect your AWS accounts. Get your first cost audit in minutes. See exactly where your money is going and how to get it back.
Try IAN on your AWS account
BYOK, no signup. Run a free audit in Claude or the CLI and get concrete findings in about a minute. See pricing when you want the agents to act.