Originally published at aicloudstrategist.com/blog/reduce-aws-bill-india-startup-playbook.html. This is a cross-post for the dev.to community.
Reduce Your Startup's AWS Bill by 40% in 30 Days: The India Founder's Playbook
By Anushka B, Founder · 2026-04-22 · 9 min read
A 40% cut in 30 days is not marketing — it's what six specific levers produce on a typical Indian SaaS startup stack that's never been optimised. This playbook itemises every lever, the INR a Series A startup should expect to save on each, and the order we execute them in.
Why 40% is the right number, not 60%
If a vendor promises 60-70% savings, they are either counting one-time FinOps automation licence offsets or quoting a pathological baseline. On a real, running Indian SaaS startup with an AWS bill of ₹8-40 lakh/month, the reproducible, audited number is 35-45% in the first 30 days, another 5-10% by day 90, and a steady-state of 50-55% below baseline if you maintain FinOps governance. We have done this 50+ times and the number is boringly consistent.
The 40% is not one big win. It is six medium wins that stack. Miss any one and you land at 25-30%. Do all six and you land at 42-48%. The discipline is in sequencing — some cuts make others harder if done out of order.
The 6 levers, in execution order
Week 1 — Orphaned resources + EBS gp3 migration (expected 6-10% cut)
Week 1 — NAT Gateway to VPC endpoints (expected 4-8% cut)
Week 2 — S3 lifecycle + intelligent tiering (expected 3-6% cut)
Week 2 — Spot for stateless + batch (expected 8-14% cut)
Week 3 — Rightsizing with real P95 (expected 6-10% cut)
Week 4 — Compute Savings Plan commitment (expected 10-15% cut)
Notice Savings Plans are last. If you commit before rightsizing you'll commit to over-provisioned instances, locking in waste for 1-3 years. We see this mistake on a third of our audits.
Week 1 lever 1: Orphaned resources and EBS gp3
The median Indian SaaS startup we audit carries 14-22% of its EBS spend on orphaned volumes and snapshots — volumes detached from terminated instances, snapshots from decommissioned RDS clusters, AMIs from a launch three CTOs ago. The fix is a one-off script plus a tag-and-sweep policy. See orphaned EBS volumes deep dive.
In parallel, migrate every gp2 volume to gp3. gp3 is 20% cheaper on the baseline and lets you buy IOPS and throughput separately. On a Series A SaaS with 8 TB of gp2, that's about ₹14,000/month saved with zero risk. Migration is an in-place modify-volume API call.
Week 1 lever 2: NAT Gateway to VPC endpoints
A single NAT Gateway in Mumbai costs ₹3.80/hour (~₹2,770/month) plus ₹3.80 per GB processed. A three-AZ prod VPC runs three of them, which is ~₹8,300/month baseline plus data charges that often dwarf the hourly cost. Every S3 GET from a private-subnet instance goes through NAT; every Secrets Manager lookup, every ECR image pull, every DynamoDB query.
Replacing those flows with VPC endpoints (S3 and DynamoDB are free gateway endpoints; most others are ~₹0.83/hour interface endpoints) typically cuts 50-70% of the NAT data-processing line item. On a ₹4 lakh/month NAT bill, that is ₹2-2.8 lakh saved. See NAT Gateway cost vs VPC endpoints.
Week 2 lever 3: S3 lifecycle and intelligent tiering
S3 Standard in Mumbai is ₹2.10/GB/month. Intelligent-Tiering is the same top price but auto-demotes cold objects to Infrequent Access (₹1.15/GB) and Archive Instant (~₹0.34/GB) at no retrieval cost. For any bucket over 1 TB you don't have a clear access pattern for, enabling Intelligent-Tiering is a no-brainer. For logs, backups, and old user uploads, a lifecycle policy to Glacier Instant Retrieval (~₹0.34/GB) or Glacier Flexible Retrieval (~₹0.19/GB) cuts storage 75-90%.
Week 2 lever 4: Spot for stateless and batch
Spot in Mumbai is typically 60-72% cheaper than On-Demand. The myth that "Spot is unreliable" comes from people running stateful DBs on it. For CI runners, EKS batch jobs, ML training, data pipelines, async workers, Karpenter or EKS managed node groups on Spot with a two-instance-family diversification policy interrupt less than 5% per month in ap-south-1. A typical SaaS stack can move 30-50% of its compute to Spot safely, which is an additional 18-30% cut on that chunk.
Week 3 lever 5: Rightsizing with real P95
AWS Compute Optimizer is a good starting point but it uses CloudWatch default metrics on a 14-day lookback. That misses weekly cycles, month-end batch, and the quarter-end reporting spike. We pull 90 days of 1-minute CPU, memory, and network metrics, compute P95 and P99, and recommend the smallest instance that carries P99 + 20% headroom.
On a 100-instance fleet, rightsizing typically reshuffles 30-45 instances one size down, saves 20-30% on those, and a few instances move up because they were actually under-provisioned. Net saving is 8-12% of the compute line.
Week 4 lever 6: Compute Savings Plan
Now, and only now, commit. A 1-year no-upfront Compute Savings Plan in Mumbai gives you ~27% off On-Demand; 3-year all-upfront gives ~52%. For an Indian startup with 18-24 months of runway, we almost always recommend 1-year no-upfront because cash is more valuable than the extra 8% discount.
Target: 75-80% coverage of your post-rightsizing steady-state compute. Leave the top 20-25% on On-Demand and Spot to absorb growth and seasonality.
What the numbers look like on a real Series A SaaS
****************
| Lever | Baseline bill impact | INR saved/month | Effort |
|---|---|---|---|
| Orphaned + gp3 | 8% | ₹1.12 lakh | 1 dev, 2 days |
| NAT -> VPC endpoints | 6% | ₹84,000 | 1 dev, 3 days |
| S3 lifecycle | 4% | ₹56,000 | 1 dev, 1 day |
| Spot migration | 11% | ₹1.54 lakh | 1 dev, 5 days |
| Rightsizing | 8% | ₹1.12 lakh | 1 dev, 4 days |
| Savings Plan | 12% | ₹1.68 lakh | CFO approval, 1 hour |
| Total | ~40-45% | ₹6.9 lakh | ~15 eng-days |
Baseline: ₹14 lakh/month. Post-playbook steady state: ₹7-8 lakh/month. Payback on the engineering time is under 3 days.
What we see in audits: the 4 "leaks" that blow the 40% plan
CloudFront price class misconfiguration — serving Indian traffic from PriceClass_All when PriceClass_100 would cover 95% of users at 30% lower egress cost.
CloudWatch Logs at default retention — never-expire retention on debug logs from a service that was sunset 14 months ago. See Cost Explorer blind spots.
Cross-region replication to a "DR" bucket that never gets read — see the cross-region egress mistake.
Data transfer between services via public endpoints — e.g., Lambda hitting RDS over its public IP instead of inside the VPC.
Frequently asked questions
Q: Will any of this cause downtime?
Only two steps carry any risk: Spot migration (stateless workloads only, with graceful-shutdown handling) and EBS gp2->gp3 (online modify; some apps see a 10-minute IOPS dip). Everything else is hot-change.
Q: We're bootstrapped, not VC-funded — does this still apply?
More so. Bootstrapped startups cannot afford 3-year commitments or complex FinOps tooling. The playbook is tuned for 1-year or shorter commitments and uses only AWS-native tools. No extra SaaS spend.
Q: Do we need to hire a FinOps person?
Not for this playbook. A mid-level DevOps engineer executes all six levers in 15 eng-days. You only need a FinOps function when your bill crosses ₹50 lakh/month, when governance becomes harder than individual optimisations.
Q: What's the risk of a 1-year Savings Plan if we pivot?
Compute Savings Plans apply to any instance family, size, OS, tenancy, or region within AWS. Even a full pivot to a different architecture usually keeps 80-90% of the commitment useful. Only a total exit from AWS strands the plan.
Q: Can we achieve this on a ₹2 lakh/month bill?
Percentages yes, absolute numbers no — the engineering time to execute is fixed. Below ₹3 lakh/month the ROI on a full playbook is marginal; focus on the top 3 levers only.
Q: How do you audit without access to our account?
We only need the Cost and Usage Report (CSV) or the last 2-3 monthly invoices (PDF). No IAM access required for the initial audit. If you later want a remediation sprint, we request a read-only cross-account role.
Q: Is this SaaS-specific or does it work for e-commerce/D2C?
The six levers apply universally. Weights change — e-commerce has more CloudFront/S3 (levers 2 and 3 matter more), AI startups have more GPU (see our AI GPU audit), fintech has RBI constraints that limit lever 6. See our SaaS, e-commerce, and D2C vertical pages.
Related reading: AWS cost audit for India · Savings Plans vs RI · FinOps assessment · AWS cost calculator
We run written 24-hour AWS cost audits. Founder-led. Free. No sales call. Send your last bill, get a PDF back. Request yours →
Top comments (0)