AWS FinOps for Vietnamese SaaS: a 90-day playbook that cuts 25–40%

AWS bills at Vietnamese SaaS companies typically run 30–60% above a healthy baseline — not because AWS is expensive, but because nobody is doing FinOps. This is a 90-day playbook that has cut 25–40% off AWS spend at Vietnamese SaaS without downgrading infrastructure.

1. Why Vietnamese SaaS typically burn AWS

• 100% on-demand because engineers fear commitments.
• Dev/staging environments never shut off on weekends.
• Legacy EBS gp2 not migrated to gp3 (a near-free 20% saving).
• S3 with no lifecycle, logs sitting forever in Standard.
• Expensive NAT Gateway egress — no VPC Endpoints for S3/DynamoDB.
• Oversized RDS, never rightsized.

2. Weeks 1–2: visibility before cutting

No tags, no FinOps. Turn on Cost Explorer at hourly + resource granularity, set up CUR export to S3, build a QuickSight dashboard or ship CUR into BigQuery/Snowflake. Enforce three mandatory tags via SCP: Owner, Environment, Service. After two weeks, at least 90% of spend must map to a specific team.

3. Weeks 3–6: quick wins, almost no dev needed

• 1-year Compute Savings Plans, no upfront. Commit 60–70% of baseline → ~27% compute saving immediately.
• EBS gp2 → gp3. Online migration, ~20% cheaper, zero downtime.
• S3 Intelligent-Tiering + lifecycle. Objects > 30 days auto-tier to Glacier Instant Retrieval.
• VPC Endpoints for S3/DynamoDB. Cuts most NAT egress.
• Auto-stop non-prod environments. Lambda + EventBridge, off 18:00–08:00 and weekends — ~65% saving on non-prod compute.
• CloudWatch Logs retention. Default Never Expire — change to 30/90 days.

4. Weeks 7–10: rightsize and selectively refactor

Run Compute Optimizer over EC2/RDS/Lambda. Most prod instances peak below 40% CPU — a one-step downsize is usually safe. Move stateless workloads to Graviton (ARM): most Node.js, Python, Java apps rebuild in 1–2 weeks for a 20–40% saving. Lambda memory tuning (via aws-lambda-power-tuning) typically cuts 15–25% per function.

5. Weeks 11–13: making it stick

• Pipe budget alerts to Slack — every team sees its weekly spend.
• Pre-deploy review: any change touching > 5% of spend needs FinOps sign-off.
• Per-team showback dashboards — accountability becomes organic.
• Quarterly rightsizing — not a one-off exercise.

6. Real results we've measured

Across six Vietnamese SaaS companies (50K–500K MAU) that ran this playbook in 2024–2026: average 32% AWS bill cut in 90 days, no production incidents, no feature cuts. The FinOps effort itself usually pays back in under six weeks.