Cloud bills have a way of growing faster than revenue. We have audited dozens of AWS accounts for Indian startups from seed-stage to Series B, and the same five mistakes appear in almost every account. The good news: all five are fixable without application changes, and together they typically reduce monthly AWS spend by 35–60%. This guide covers what to look for and exactly what to do — applicable whether you are already on cloud or planning a migration.
Mistake 1: Over-Provisioned EC2 Instances
Most teams select an instance type during initial setup and never revisit it. AWS Compute Optimizer analyses actual CPU, memory, and network utilisation over 14 days and recommends right-sized alternatives for free. In our audits, 60–70% of EC2 instances run below 20% average CPU utilisation and can be downsized by one or two instance families with no perceptible performance impact. Savings: 30–50% on your compute line item — typically the largest single component of an AWS bill.
Mistake 2: No S3 Lifecycle Policies
Object storage accumulates silently. Application logs, database backups, build artefacts, and media uploads from months past sit in Standard storage when they have not been accessed in ages. Set explicit lifecycle rules using the S3 storage classes framework: transition to S3-IA after 30 days, Glacier after 90 days, delete after 365. Alternatively, S3 Intelligent-Tiering automates this. On accounts with large S3 buckets, this single change commonly saves ₹30,000–₹2 lakh monthly.
Mistakes 3 & 4: Dev/Test Running 24/7 and Orphaned Storage
Non-production environments running continuously cost the same as production but are only used 8–10 hours a day, 5 days a week. The AWS Instance Scheduler (free) automatically stops and starts instances on a schedule. Stopping dev/test outside working hours and weekends eliminates approximately 65% of their runtime cost. Separately: when an EC2 instance is terminated, its EBS volumes often remain. We routinely find 10–30% of EBS spend going to orphaned volumes from terminated instances. A weekly Lambda function that detects unattached volumes and sends a Slack alert is enough to stop the accumulation.
Mistake 5: No Reserved Capacity Planning
On-demand EC2 pricing is 2–4x more expensive than Savings Plans for predictable workloads. Compute Savings Plans are flexible across instance families and regions — a 1-year, no-upfront-payment Savings Plan pays for itself within 3–4 months. Most startups should run at least their production database and application tier on Savings Plans. This is especially relevant if you are undertaking a cloud architecture review — getting compute commitments right from the start saves significantly over ad hoc corrections later.
Taking Action
Priority order: (1) Run Compute Optimizer — free, no risk. (2) Add S3 lifecycle policies. (3) Schedule dev/test shutdowns. (4) Audit and clean orphaned EBS volumes. (5) Commit to Savings Plans based on your right-sized baseline. Final step: enable AWS Cost Anomaly Detection with Slack notifications so you are alerted within 24 hours if spend spikes unexpectedly. Crenosoft's DevOps team audits AWS accounts and implements all five fixes — on a typical Series A account, we recover ₹1–8 lakhs per month in wasted spend. Book a free cost audit with our team.