Skip to content
Back to case studies
SanitizedCost Optimization2024

Cloud Cost Optimization

Combined blind spend cut up to 55%; idle EC2 eliminated; SP coverage from 0% to 78% where it fit.

Role
Cloud Consultant
Context
Fintech / E-Commerce / SaaS
Duration
2 weeks per client (6 weeks total)
Team
Solo consultant

Tech stack

  • AWS Cost Explorer
  • Trusted Advisor
  • Python
  • Terraform
  • CloudWatch
  • S3
  • Lambda
  • SNS
  • QuickSight
  • AWS Budgets
Executive summary

Business problem and production context.

Business problem

Three enterprise clients were overspending on AWS. One had significant idle EC2 inventory, another was running oversized RDS instances at low utilization, and the third had no Savings Plans despite predictable workloads, all with no visibility into per-team or per-service costs.

Architecture decision

Conducted 2-week architecture audits per client using AWS Cost Explorer, Trusted Advisor, and custom Python scripts. Implemented resource right-sizing (EC2, RDS, ElastiCache), converted qualifying workloads to Savings Plans and Reserved Instances, migrated batch jobs to Spot/Fargate Spot, enforced S3 lifecycle policies, and built FinOps dashboards with cost allocation tags and per-team budgets with SNS alerts.

awsAWS Account / FinOps programRequestResponseCOST SURFACEOPTIMIZATION WORKFLOWCOMMITMENT + REPORTING12345678Cost Explorerby tag + serviceTrusted Advisoridle + right-sizeCUR Athenaline-item drillASGright-sized fleetsLambdaschedule on/offCompute Optimizerinstance matchSavings Plans78% coverageReserved Inststable baseloadQuickSightexec dashboardCROSS-CUTTING: Budget alerts per team · Tag enforcement SCP · Idle resource auto-stop · Monthly variance review
What Rahul owned

The useful proof is the decision surface, not only the result number.

These are the parts of the work that show production judgment: ownership, constraints, rollback, cost, and observability.

Owned scope

  • Multi-client AWS cost audit narrative with billing analysis, right-sizing, lifecycle policies, and FinOps automation
  • Savings methodology that separates quick cuts from durable operating changes
  • Cost visibility model for engineering and finance to discuss the same drivers

Evidence artifacts

Billing analysisRight-sizingSavings Plans reviewLifecycle policies₹50L+/yr sanitized narrative
Constraints

What would fail first?

Every project has constraints. The useful work is naming them early enough that rollback, cost, and ownership are designed before an incident.

Mixed workloads

Savings came from different levers across compute, storage, databases, logs, and commitment strategy.

Business risk

Cutting cost could not reduce reliability or remove capacity required for real traffic.

Proof framing

Client details are sanitized, so savings need clear methodology and NDA-safe context.

Decision table

Trade-offs accepted on purpose.

Option
Decision
Reason
Savings Plans vs Reserved Instances
Savings Plans for compute flexibility across instance families; RIs only for stable RDS workloads
Savings Plans vs Reserved Instances: Savings Plans for compute flexibility across instance families; RIs only for stable RDS workloads.
Spot vs On-Demand for batch
Fargate Spot with graceful interruption handling
Spot vs On-Demand for batch: Fargate Spot with graceful interruption handling. 70% cost savings with <2% interruption rate.
Custom dashboards vs AWS native
QuickSight for executive dashboards, CloudWatch for engineering
Custom dashboards vs AWS native: QuickSight for executive dashboards, CloudWatch for engineering. Different audiences need different views.
Aggressive vs conservative right-sizing
Used 30-day P95 metrics with 20% headroom, balancing cost with performance safety margin
Aggressive vs conservative right-sizing: Used 30-day P95 metrics with 20% headroom, balancing cost with performance safety margin.
Risk table

Production risks and how they were controlled.

Risk
Mitigation
Owner
Cutting useful capacity
Separated idle waste from traffic-protecting capacity before recommending changes.
Engineering + finance
Savings disappeared next month
Added tagging, owner review, budgets, and recurring cost-driver checks.
Platform owner
Commitments locked in the wrong shape
Reviewed workload stability before Savings Plans or reserved capacity.
Finance + platform
Implementation approach

Small steps, visible changes, fewer surprises.

  1. 01

    Ran 3-day discovery per client: extracted Cost Explorer data, Trusted Advisor reports, and CloudWatch utilization metrics across all regions.

  2. 02

    Built Python analysis scripts to identify idle resources (EC2 at <5% CPU, RDS at <8% utilization, unattached EBS volumes, unused Elastic IPs).

  3. 03

    Modeled Savings Plan scenarios using AWS Cost Explorer APIs, comparing 1-year vs 3-year, All Upfront vs No Upfront, across Compute and EC2 families.

  4. 04

    Implemented automated enforcement: S3 lifecycle policies (30-day IA, 90-day Glacier), Lambda-based schedulers for dev environment shutdown, and budget alerts at 80/100% thresholds.

  5. 05

    Built QuickSight dashboards with cost allocation by team, service, and environment, giving CFOs and engineering leads visibility they never had.

Result

Before and after.

Combined AWS Spend
baseline
55% lower
Idle EC2 Instances
47
0
Cost Visibility
Monthly PDF from finance
Real-time per-team dashboards
Savings Plan Coverage
0%
78% of eligible compute

Rollback path

High-risk cost changes were staged, monitored, and reversible; lifecycle and schedule-based changes were kept explicit.

Cost considerations

The public story stays sanitized: ₹50L+/yr client savings and selected 30-40% reductions where details can be shared.

Observability notes

Useful signals were cost by owner, unit cost, anomaly alerts, idle inventory, and post-change service health.

Related services

Forward Deployed AI EngineeringAWS Production Architecture ReviewGenAI / RAG Production Readiness

Have a similar production problem?

Send the current architecture, bill concern, deployment pain, or GenAI reliability issue. I will help you find the first real bottleneck.