Cloud Cost Optimisation in 30 Days: A Realistic Plan

30 days is enough to cut a typical cloud bill by 30-50% if you sequence the work correctly. Here's the day-by-day plan I run on every engagement.

By Andrii Votiakov on 2026-04-20

Cost optimisation as an open-ended project never ends. As a 30-day sprint with a defined sequence, it usually delivers 30-50% savings without much disruption. Here's the playbook.

Quick answer

Days 1-7: visibility and quick wins (kill zombies, set retention, fix obvious waste). Days 8-14: right-size compute and storage. Days 15-21: lock in commitment discounts and migrate where it pays. Days 22-30: guardrails to keep it cheap. By day 30: 30-50% off the bill, sustainable, with monitoring in place.

Days 1-3: Get the picture

Goal: understand where the money goes.

• Pull a 90-day Cost Explorer / Billing report grouped by service, then by linked account, then by tag
• Identify the top 5 services by spend (usually compute, database, network, storage, observability)
• For each top service, identify the top 5 resources (e.g., the 5 biggest EC2 instances, 5 biggest RDS, 5 biggest S3 buckets)
• Note any "AWS Cost" / "Other" / "Tax" line items — they'll often hide things like data transfer

You should be able to fit the result on a single page. If you can't, you don't have visibility yet.

Days 4-7: Quick wins

Goal: free money. No risk, no architecture changes.

• Delete unattached EBS volumes and old snapshots
• Delete unused Elastic IPs / static IPs / load balancers
• Set retention on log groups (CloudWatch / Stackdriver / Azure Monitor)
• Stop dev/staging instances overnight and weekends (Instance Scheduler or a 5-line Lambda)
• Delete unused load balancers, idle NAT Gateways in test accounts, orphan TGW attachments
• Review Cost Anomaly Detection findings — usually there's a 2-month-old anomaly nobody actioned

This usually takes 5-10% off the monthly bill in 4 days.

Days 8-10: Right-size compute

Goal: pay for what you actually use.

• Pull 14 days of CPU + memory metrics for every compute resource
• Apply right-sizing rules (see the EC2 post)
• For containers: audit Kubernetes requests/limits, tighten where actual usage is far below request
• Roll changes one workload at a time, monitor for issues

Typical saving: another 15-25% off compute.

Days 11-14: Right-size storage and observability

• S3 / Cloud Storage / Blob lifecycle rules (see the S3 post)
• Database storage type (gp3 over gp2, smaller IOPS provisioning where overprovisioned)
• CloudWatch / Datadog cardinality cleanup
• Log volume reduction (drop DEBUG, drop health checks, sample 200s)

Saves another 10-15% on monitoring/storage tail.

Days 15-18: Migrate where it pays

These are bigger changes; only do them if confidence is high.

• Graviton / ARM migration for compatible workloads (see the Graviton post)
• VPC endpoints for S3, DynamoDB, ECR, Logs to cut NAT processing
• CloudFront / CDN in front of any high-egress workload
• Spot instances for stateless tier (with proper interruption handling)

Each of these is days, not hours, but ROI is high and lasting.

Days 19-21: Commitment discounts

Goal: lock in the savings.

• Buy 1-year Compute Savings Plans / CUDs on the new floor (after right-sizing!)
• Buy RDS / Cloud SQL / Azure SQL reservations on stable databases
• Don't 3-year-commit unless usage is predictably stable for 3 years

Saves another 25-40% on the steady-state portion that's commitment-eligible.

Days 22-25: Tagging and chargeback

• Tag every resource with team, service, env (mandatory)
• Set up a weekly cost-by-team report
• Identify the team with the biggest "untagged" line — usually 20-40% of spend
• Run an hour-long session per team showing them their costs

Doesn't directly save money on day one, but shifts behaviour for the next quarter.

Days 26-28: Guardrails

Goal: stop bills from growing back.

• Budgets with alerts at 50/80/100% per linked account
• Anomaly detection turned on
• Resource monitor (Snowflake) / per-user query limit (BigQuery)
• IAM policy restricting expensive instance types in non-prod
• Reserved/RI/SP coverage report scheduled weekly

Days 29-30: Document and hand off

• Write down what was changed, why, and the rollback path
• Create a "monthly review" checklist for ongoing maintenance
• Schedule the next review in 90 days
• Hand off the dashboards (cost-by-tag, RI/SP coverage, anomalies) to the team

What this looks like on a real bill

Recent client (~$42k/month starting):

Final: $22,500/month, ~46% reduction in 30 calendar days.

What goes wrong

Common pitfalls:

• Trying to do everything in week 1. The sequence matters — don't buy commitment discounts before right-sizing.
• No tagging before chargeback. You can't charge back what you can't measure.
• Skipping guardrails. Without budgets and anomaly detection, you'll be back here in 6 months.
• Trying to right-size everything yourself. Use Compute Optimizer / Recommender first, fight the recommendations, but don't reinvent the math.
• Forgetting Datadog / observability. It's often 15-25% of total cloud spend and gets ignored.

If you'd like me to run this 30-day plan on your account on a pay-for-savings basis, book a call. 30-minute conversation, no pitch.