$47,000 a month down to $28,000 a month. That is a 40% cut to a SoCal SaaS company's AWS bill, or $228,000 in annualized savings, delivered in six weeks without touching a single product feature. The client, a B2B SaaS company headquartered in Irvine, had watched their bill climb from $15,000 to $47,000 over 18 months while revenue only doubled — infrastructure costs had grown 3x against 2x revenue growth, and nobody could explain the gap.
They had already tried the usual approaches — reserved instances, some right-sizing, shutting down dev environments at night. It barely moved the needle. The real waste was hiding in places that required deep engineering analysis to find. Here is how we got to the $19,000/month reduction, worked backward from the number.
Where the $19,000 Came From
An AI-powered audit in Week 1 located the waste. We fed their AWS Cost Explorer data, CloudWatch metrics, and infrastructure-as-code configs into an analysis pipeline. Using Claude, we generated a comprehensive resource utilization map that would have taken a human engineer days to compile. Three categories of waste accounted for 85% of the excess spend.
Oversized RDS instances. Their production database ran on a db.r6g.4xlarge ($2,800/month) but averaged 12% CPU utilization. The AI correlated CloudWatch metrics with query logs and determined that a db.r6g.xlarge ($700/month) could handle their workload with headroom.
Orphaned resources. 23 EBS volumes from terminated instances, 8 unused Elastic IPs, 4 load balancers pointing to empty target groups, and an entire staging environment that hadn't been accessed in five months. Total: $3,200/month for resources doing absolutely nothing.
Unoptimized data transfer. Their application made cross-AZ API calls for every request, doubling their data transfer costs. A simple architecture change — colocating services in the same AZ — would save $4,500/month.
How the Cuts Got Made
Weeks 2 through 4 were where our Vietnam pod earned its keep. Every optimization was implemented, tested, and validated during overnight hours — zero disruption to the US engineering team's sprint work.
Our engineers worked through the changes methodically. Database migration to smaller instances happened over a weekend, with our overnight team monitoring performance metrics in real time and ready to roll back if latency spiked. It didn't.
The orphaned resource cleanup was straightforward but tedious — exactly the kind of work that gets deprioritized when US engineers have feature work to deliver. Our team cleared it in two nights.
The cross-AZ optimization required careful service reorganization. Our engineers mapped every service dependency, tested the new topology in staging, and executed the migration in three phases across three nights.
Keeping the Number Down
Cutting costs once is easy. Keeping them down is the hard part, and it is why the $28,000 has held steady for four months. In Weeks 5 and 6 we built automated guardrails using AI-assisted monitoring:
A daily cost anomaly detector that alerts when any service's spend deviates more than 15% from its 30-day average. The AI baseline adjusts automatically for known patterns like month-end processing spikes.
An automated right-sizing recommender that runs weekly, analyzes instance utilization, and creates pre-approved change requests for the US team to review each Monday morning.
A resource lifecycle policy that automatically tags unused resources after 14 days and terminates them after 30 days — with Slack notifications at each stage.
The client's infrastructure now scales linearly with their business instead of exponentially. Our engagement cost was a fraction of the $228,000 saved. Cloud waste isn't a technology problem — it's an attention problem. Your US team is busy building features. Our overnight team makes sure the infrastructure stays efficient while they sleep.