7 AWS Changes That Cut Our Client’s Bill by 40%

When a SaaS startup came to us with a $4,200/month AWS bill and 18 months of runway left, we had a clear mandate: cut costs without breaking anything. Six weeks later, their bill was $2,450 — a 42% reduction — and response times actually improved.

Here’s exactly what we did.

The Starting Point

Before touching anything, we spent a week auditing. The AWS Cost Explorer and Trusted Advisor findings were sobering:

60% of EC2 instances running at less than 10% CPU utilization
Three RDS instances with no connections after 6pm
S3 buckets storing 800GB of access logs from 2021
No Reserved Instances despite running the same workload for 14 months
CloudWatch Log retention set to “Never expire”

The biggest wins were obvious. Here’s how we tackled them in order of impact.

1. Right-Size EC2 Instances

The biggest line item was compute: $1,800/month across 12 instances. CloudWatch metrics told the real story — these instances were massively overprovisioned.

We used AWS Compute Optimizer to get specific recommendations, then tested each change in staging before applying to production:

6x t3.large → t3.medium (saved $180/month)
2x m5.xlarge → m5.large (saved $210/month)
1x r5.2xlarge → r5.xlarge (saved $140/month)

Total savings: ~$530/month

The r5 instance was running a memory-intensive process we moved to Lambda, which also improved latency.

2. Purchase Reserved Instances

The client had been running the same production workload on On-Demand instances for 14 months. We purchased 1-year Standard Reserved Instances for the 3 instances we knew would run 24/7.

Reserved Instances offer 30-40% discount over On-Demand with 1-year commitment, 60%+ with 3-year.

Total savings: ~$320/month

3. Implement S3 Intelligent Tiering + Lifecycle Policies

The S3 audit revealed three issues:

800GB of old log files with no lifecycle policy
No tiering on infrequently accessed data
Versioning enabled but no expiration on old versions

We added lifecycle rules to:

Delete objects older than 90 days in the logs bucket
Move objects to S3 Standard-IA after 30 days in the data bucket
Expire non-current versions after 30 days

aws s3api put-bucket-lifecycle-configuration \
  --bucket my-data-bucket \
  --lifecycle-configuration file://lifecycle.json

Total savings: ~$95/month

4. Schedule Non-Production Instances

The staging environment ran 24/7, but engineers only used it 9am-6pm weekdays. We used AWS Instance Scheduler to automatically stop instances outside business hours.

The math: 12 hours off per weekday + full weekends = ~65% of the week idle.

Total savings: ~$180/month

5. Optimize RDS

Three RDS instances needed attention:

Multi-AZ for dev: Dev and staging didn’t need Multi-AZ. Disabling it cut the database cost in half for those environments.

Read replica right-sizing: A db.r5.large read replica handling only reporting queries — replaced with db.t3.medium.

Aurora Serverless v2: The reporting database had spiky usage (heavy on Monday mornings, quiet the rest of the week). Aurora Serverless v2 scaled to zero during off-hours.

Total savings: ~$210/month

6. Clean Up Unused Resources

A systematic audit found:

8 unattached EBS volumes (snapshots kept, volumes deleted)
3 unused Elastic IPs ($3.60/month each, but the principle matters)
12 old AMIs and their associated snapshots
4 Elastic Load Balancers with no targets

This sounds small, but unattended resources compound. Setting up monthly AWS Config rules to flag orphaned resources prevents future drift.

Total savings: ~$45/month

7. Fix CloudWatch Log Retention

CloudWatch Logs with “Never expire” retention is a slow leak. We audited all log groups and set appropriate retention:

Application logs: 30 days
Access logs: 7 days
Audit/compliance logs: 365 days
Debug logs: 3 days

Total savings: ~$70/month

The Result

Change	Monthly Savings
EC2 right-sizing	$530
Reserved Instances	$320
RDS optimization	$210
Non-prod scheduling	$180
S3 lifecycle policies	$95
CloudWatch log retention	$70
Unused resource cleanup	$45
Total	$1,450/month

From $4,200 to $2,750/month — and we found additional savings in month two by applying the same patterns to their data pipeline.

What You Should Do This Week

Open Cost Explorer → set to last 3 months → sort by service
Run Trusted Advisor (Business/Enterprise Support required for full checks)
Check EC2 Compute Optimizer for rightsizing recommendations
Audit your S3 buckets for lifecycle policies

Most AWS bills have 20-30% waste. The question is just whether you find it before your runrate does.

Want a free audit of your AWS infrastructure? Get in touch — we’ll identify your top 5 cost reduction opportunities at no charge.

7 AWS Changes That Cut Our Client's Bill by 40%