A rapidly growing FinTech SaaS startup in India saw their monthly AWS bill balloon from $2,000 (₹1.6L) to over $8,500 (₹7L) in under 8 months as user acquisition skyrocketed. The engineering team was provisioning massive instances to avoid downtime, leading to extreme over-provisioning and skyrocketing On-Demand compute costs.
When the startup landed a major news feature, their traffic spiked by 300%. To handle the load, engineers manually launched massive `m5.4xlarge` and `c5.4xlarge` instances and just left them running 24/7. Since traffic was only high during daytime business hours, these massive servers sat completely idle for 14 hours a day, burning cash at premium On-Demand rates.
By analyzing CloudWatch metrics, the team discovered that their absolute minimum baseline load (the databases, core background workers, and minimum web nodes) required about `m5.2xlarge` worth of continuous capacity. They purchased a 1-year No Upfront Standard Reserved Instance (RI) for this capacity. This instantly slashed the cost of their 24/7 workloads by ~40%.
Instead of manually provisioning massive servers during the day and forgetting to turn them off at night, the team configured AWS Auto Scaling Groups (ASG). The ASG was set to monitor CPU utilization. When daytime traffic hit, the ASG automatically spun up smaller `c5.large` instances to handle the load, and systematically terminated them when traffic fell back to the baseline in the evening.
The startup had a massive nightly batch processing job that reconciled financial transactions. This job previously ran on expensive On-Demand instances. They moved this architecture to AWS Spot Instances. Because the batch job was fault-tolerant and could resume if interrupted, Spot Instances allowed them to bid for unused AWS capacity at a 90% discount compared to On-Demand.
By implementing a blended approach of Reserved Instances for the baseline, Auto Scaling for the dynamic load, and Spot Instances for batch processing, the startup reduced their monthly AWS bill from $8,500 to $2,650—an annual saving of roughly $70,000 (₹45+ Lakhs).
Not only did they save massive amounts of capital, but the introduction of Auto Scaling actually made their infrastructure more resilient than the previous manual over-provisioning method.