Cloud Cost Audit

# Cloud Cost Optimization Audit

Analyze cloud infrastructure spend across AWS, Azure, and GCP. Identify waste, rightsizing opportunities, and reserved instance savings.

What This Skill Does

When given cloud spend data (billing exports, cost explorer screenshots, or manual input), this skill:

1. Categorizes spend across 8 cost domains (compute, storage, networking, databases, AI/ML, observability, security, licensing)
2. Identifies waste patterns using 12 common anti-patterns
3. Calculates savings with specific dollar amounts per optimization
4. Prioritizes actions by effort vs. impact (quick wins → strategic moves)
5. Generates executive summary with 90-day roadmap

Cost Domains & Benchmarks (2026)

1. Compute (typically 40-55% of total) - **Idle instances**: >30% idle = waste. Benchmark: <10% idle capacity - **Rightsizing**: 60% of instances are oversized by 1+ size category - **Spot/preemptible**: Batch workloads not on spot = 60-80% overpay - **Reserved/savings plans**: On-demand for steady-state = 30-50% overpay - **Container density**: <40% CPU utilization on nodes = poor bin-packing

2. Storage (typically 10-20%) - **Tiering**: Data not accessed in 90 days still on hot storage = 60-80% overpay - **Snapshot sprawl**: Orphaned snapshots older than 30 days - **Duplicate data**: Cross-region replication without business justification - **Object lifecycle**: No lifecycle policies = guaranteed bloat

3. Networking (typically 8-15%) - **Cross-AZ traffic**: Unnecessary data transfer between zones ($0.01-0.02/GB) - **NAT gateway abuse**: High-throughput through NAT vs. VPC endpoints - **CDN miss rate**: >20% miss rate = CDN config issue - **Egress optimization**: No committed use discounts on egress

4. Databases (typically 10-20%) - **Over-provisioned RDS/Cloud SQL**: Multi-AZ for dev/staging environments - **Read replica sprawl**: Replicas with <5% query load - **DynamoDB/Cosmos over-provisioning**: Provisioned capacity 3x+ actual usage - **License waste**: Commercial DB when open-source works

5. AI/ML Infrastructure (growing — 5-25%) - **GPU idle time**: Training instances running 24/7 for 4hr/day workloads - **Inference over-provisioning**: GPU instances for CPU-viable inference - **Model storage**: Old model versions consuming storage - **API costs**: Frontier model API calls without caching layer

6. Observability (typically 3-8%) - **Log ingestion bloat**: Debug logs in production, duplicate log streams - **Metric cardinality**: High-cardinality custom metrics ($$$) - **Trace sampling**: 100% trace sampling when 10% suffices - **Retention overkill**: 13-month retention for non-compliance data

7. Security (typically 2-5%) - **WAF rule bloat**: Managed rule groups not actively tuned - **Key management**: KMS keys for non-sensitive data - **Compliance scanning**: Overlapping tools doing same checks

8. Licensing (typically 5-15%) - **Shelfware**: Paid seats not logged in 60+ days - **Duplicate tools**: Multiple tools solving same problem - **Enterprise tiers**: Enterprise features unused, paying enterprise price

12 Waste Anti-Patterns

| # | Pattern | Typical Waste | Fix Effort | |---|---------|--------------|------------| | 1 | Zombie resources (stopped but attached) | 5-15% of bill | Low | | 2 | Over-provisioned instances | 15-30% compute | Medium | | 3 | No reserved capacity strategy | 25-40% compute | Medium | | 4 | Hot storage hoarding | 40-70% storage | Low | | 5 | Cross-AZ data transfer abuse | 10-30% network | Medium | | 6 | Dev/staging mirrors production | 20-40% of envs | Low | | 7 | Orphaned snapshots/AMIs | 3-8% storage | Low | | 8 | Log ingestion without sampling | 30-60% observability | Low | | 9 | GPU instances for CPU workloads | 70-85% compute | Medium | | 10 | No spot/preemptible for batch | 60-80% batch | Medium | | 11 | Shelfware licenses | 20-40% licensing | Low | | 12 | No tagging = no accountability | Unmeasurable | High |

Savings Estimation Framework

For each finding, calculate: ``` Annual Savings = (Current Cost - Optimized Cost) × 12 Implementation Cost = Engineering Hours × Loaded Rate ROI = (Annual Savings - Implementation Cost) / Implementation Cost Payback Period = Implementation Cost / (Annual Savings / 12) ```

Typical Savings by Company Size | Company Size | Monthly Cloud Spend | Typical Waste % | Annual Savings | |-------------|-------------------|----------------|---------------| | Startup (5-15) | $2K-$15K | 35-50% | $8K-$90K | | Growth (15-50) | $15K-$80K | 25-40% | $45K-$384K | | Mid-market (50-200) | $80K-$500K | 20-35% | $192K-$2.1M | | Enterprise (200+) | $500K-$5M+ | 15-25% | $900K-$15M+ |

Output Format

1. Generate a report with:
2. Executive Summary: Total spend, waste identified, savings potential, top 3 quick wins
3. Domain Breakdown: Spend per domain vs. benchmarks
4. Findings Table: Each finding with current cost, optimized cost, savings, effort, priority
5. 90-Day Roadmap: Week 1-2 quick wins, Week 3-6 medium effort, Week 7-12 strategic
6. Governance Recommendations: Tagging strategy, budget alerts, review cadence

Usage

• Provide your cloud billing data in any format:
• AWS Cost Explorer export / Azure Cost Management / GCP Billing
• Monthly bill summary
• Architecture description with approximate sizing
• Or just describe your stack and team size for estimates

The agent will analyze and produce the full optimization report.

---

Want Industry-Specific Cloud Optimization?

Different industries have different compliance, data residency, and workload patterns that change the optimization calculus entirely.

Get your industry context pack — pre-built frameworks for Fintech, Healthcare, Legal, SaaS, Ecommerce, Construction, Real Estate, Recruitment, Manufacturing, and Professional Services.

🛒 Browse packs: https://afrexai-cto.github.io/context-packs/ 🧮 Calculate your AI savings: https://afrexai-cto.github.io/ai-revenue-calculator/ 🤖 Set up your agent: https://afrexai-cto.github.io/agent-setup/

• Bundle deals:
• Pick 3 packs: $97
• All 10 packs: $197
• Everything bundle: $247