When a $30M-revenue B2B SaaS company saw its AWS bill hit $240,000 a month—and gross margins dip below 60%—leadership didn’t panic. They called PADISO for a no-holds-barred cloud cost teardown. The result? A 40% reduction, to $144,000 monthly, with no performance trade-offs and a clear path to unit-cost-aware growth. Here’s exactly how we did it, step by step, so you can spot the same leaks in your own environment.
Table of Contents
- The 40% Cloud Cost Teardown: Overview
- The Starting Point: Where SaaS Cloud Bills Bleed
- Phase 1: Visibility & FinOps Discipline
- Phase 2: Right-Sizing Compute and Storage
- Phase 3: Architecture & Reserved Instances
- Phase 4: Data Transfer & Network Optimizations
- Phase 5: Embedding Continuous Cost Optimization
- AI-Accelerated Cost Optimization: The Next Frontier
- Summary & Your Next Steps
- How PADISO Helps B2B SaaS Leaders Own Their Cloud Economics
The 40% Cloud Cost Teardown: Overview
The company, a B2B SaaS platform serving mid-market logistics providers, had grown quickly on AWS. Yet its infrastructure costs were growing faster than revenue. The CFO projected that at current run rates, cloud spend would consume over 40% of revenue within two years. The board needed action.
We executed a 12-week engagement that combined fractional CTO oversight with hands-on platform re-architecture. The savings broke down into five categories: visibility and FinOps discipline (8% of total savings), right-sizing compute and storage (35%), architecture changes and reserved instance purchases (40%), data transfer and network optimization (10%), and ongoing governance (7%). None of these moves required a rewrite of the application; they were surgical, evidence-based, and immediately accretive to EBITDA.
Engaging a fractional CTO — as we’ve done for firms from Gold Coast to Ottawa — was the first step, because lasting cloud economics demand leadership, not just tooling. We embedded that leadership to make sure the savings stuck.
The Starting Point: Where SaaS Cloud Bills Bleed
Before any cost-cutting, we mapped the entire AWS estate: 14 accounts, over 1,200 EC2 instances, 8 Elastic Load Balancers, 12 RDS databases, and data flowing across three regions. The biggest line-items were EC2 (48%), RDS (18%), data transfer (12%), and S3 (8%).
As the AWS cost optimization best practices highlight, without granular tagging and monitoring, waste is invisible. And indeed, only a third of resources had meaningful tags; the rest were “untagged” noise. We couldn’t even tell which microservice was driving the most spend.
The invoice deep-dive
Using AWS Cost Explorer and a third-party FinOps platform, we attributed 60% of costs to the production environment, 25% to staging/testing, and 15% to idle or forgotten resources. A deep-dive into EC2 usage showed that 35% of instances averaged <15% CPU utilization over a 30-day window. Several RDS instances were running on provisioned IOPS that far exceeded their workload requirements. Data transfer between VPCs in the same region — simply due to misconfigured peering — was costing $9,000 per month.
This initial diagnosis took less than two weeks and provided the authoritative blueprint for every subsequent move.
Phase 1: Visibility & FinOps Discipline
We couldn’t optimize what we couldn’t measure. The first phase established a FinOps foundation that turned the monthly invoice from a shock into a management tool.
Tagging and cost allocation
We deployed a mandatory tagging policy — owner, environment, service, and cost center — and backfilled tags for all running resources. A dashboard showing cost per microservice, per team, gave engineering leaders real-time visibility. For a platform engineering team in San Francisco, similar tagging clarity has surfaced thousands per month in orphaned resources.
This alone didn’t cut spend, but it ended the blame game and created cost-aware teams. Within 30 days, every sprint review included a “cost per tenant” metric.
Budget alerts and anomaly detection
We configured AWS Budgets to send alerts at 80%, 100%, and 120% of forecasted spend per service. These alerts caught a forgotten RDS instance in a sandbox account that was running $1,400/month with zero connections — a three-second fix. We also set up anomaly detection on CloudWatch metrics, which would later flag a spike in NAT gateway traffic (Phase 4).
The SaaS Cloud Cost Optimization Playbook emphasizes that this “visibility foundation” is the single highest-return investment a SaaS company can make — and we saw the payback in weeks, not months.
Phase 2: Right-Sizing Compute and Storage
With clear cost attribution, we targeted the low-hanging fruit: compute and storage that were grossly over-provisioned.
EC2 instance right-sizing
We moved 40% of production workloads from m5.large to c6g.large (Graviton2) instances with zero code changes, achieving a 20% performance lift and a 25% cost reduction. Non-critical batch jobs were shifted to spot instances, cutting their cost by 70%. For a Seattle SaaS company we advised, a similar graviton migration cut compute spend by nearly a third in a single weekend.
We also rightsized dozens of t3.medium instances that had been provisioned but rarely used above baseline; memory profiling proved they could run on t3.small for a 40% savings.
Container optimization
The team’s EKS cluster had pods requesting far more memory than they consumed. Adjusting request/limit ratios to match actual usage freed capacity, allowing us to reduce node count by 15%. We introduced vertical pod autoscaling (VPA) in recommendation mode to guide future settings. Combined with spot instances for dev environments, container costs dropped 28%.
Storage tiering
S3 buckets held 400 TB of data, much of it logs and backups never accessed after 30 days. We implemented lifecycle policies to transition objects to S3-IA after 30 days and Glacier Deep Archive after 90. This alone cut S3 spend by 55%. EBS volumes were migrated from gp2 to gp3 for the same or better performance at a 20% lower cost per GB. We also deleted over 300 unattached EBS volumes that had been accumulating since the company’s early days.
The CloudZero SaaS cost optimization guide notes that storage often hides the “latent waste” that older SaaS companies carry. Our client was no exception.
Phase 3: Architecture & Reserved Instances
Right-sizing is tactical; architecture is strategic. Here we locked in a durable cost structure.
Reserved Instances and Savings Plans
We analyzed the steady-state workloads and purchased 1-year all-upfront Reserved Instances for 60% of them, slashing those EC2 costs by 40%. For the remaining 40% of variable workloads, we applied Compute Savings Plans, which gave flexibility across instance families and regions while still offering a 28% discount. The blended effective EC2 rate dropped from $0.085/hr to $0.048/hr.
Eliminating idle resources
We consolidated 8 load balancers into 4 by merging host-based routing rules; the ALB fees alone saved $2,100/month. Non-production environments were scheduled to shut down from 8 PM to 6 AM on weekdays and all weekend via Lambda scripts. The combined idle-resource cleanup freed another 7% of total spend.
Database optimization
RDS instances were right-sized based on CloudWatch metrics; one PostgreSQL instance running on db.r5.2xlarge had an average CPU of 9% and was throttling due to provisioned IOPS limits. Moving it to db.r6g.large with the same IOPS (but using gp3 storage) dropped that database’s cost by 52% with no latency change. For a Dallas logistics platform, we saw similar RDS savings after moving to Aurora Serverless for variable workloads, making the per-query cost track usage instead of peak provisioning.
We also optimized read replicas: two replicas for analytics were only queried during business hours, so we scaled them down outside those windows.
Phase 4: Data Transfer & Network Optimizations
Data transfer costs are notoriously opaque. This client had $10,000/month in cross-AZ and NAT gateway charges—typical for a growing multi-tenant platform.
Data transfer charges demystified
Cross-AZ traffic is free within an AZ but costs $0.01/GB between AZs. Many microservices were randomly spread across all three AZs, generating substantial inter-AZ chatter. We reconfigured placement groups to keep latency-critical services in the same AZ where possible, and used VPC endpoints where appropriate. This cut cross-AZ costs by 35%.
A fintech in New York had a similar challenge: we re-architected their VPC footprint to minimize inter-AZ data transfer, saving $15,000/month without adding latency. The Oracle cloud cost optimization guide reinforces that network egress is one of the top three hidden costs in the public cloud.
VPC peering and transit gateways
The client had redundant VPC peering connections — some even circular — that drove unnecessary data processing fees. We collapsed them into a hub-and-spoke model using AWS Transit Gateway, trimming data processing charges by 22%. For Miami companies handling cross-border data, this type of cleanup is essential to maintain SOC 2-ready architecture while controlling cost.
NAT gateway processing was $2,300/month; simply rerouting internal traffic through VPC endpoints for S3 and DynamoDB reduced that to $600/month.
Phase 5: Embedding Continuous Cost Optimization
One-time savings evaporate unless the organization develops a cost-conscious muscle memory. The final phase locked in the gains.
Building a FinOps culture
We established a weekly “cloud cost stand-up” attended by the CTO, lead engineers, and the finance team. Each team reviewed their cost-per-unit metric and proposed one optimization per sprint. Over the following quarter, these small, continuous improvements sustained the 40% reduction and even squeezed out another 3%.
A Jacobian Engineering whitepaper underscores that governance is the decisive factor: “The organizations that sustain cloud cost savings are those that treat FinOps as a cross-functional discipline, not a one-time project.” Our engagement modeled exactly that.
Automating cost anomaly detection
We built a lightweight Lambda function that scans Cost Explorer daily and posts Slack alerts if any service line exceeds three standard deviations from the 30-day rolling average. This caught a runaway Glue job that was spinning up hundreds of DPUs — a bug that would have cost $8,000 had it run over a weekend.
We also automated non-production shutdowns and tag compliance checks using AWS Config rules. For a mid-market operator we work with in Austin, similar automation has prevented five-figure surprises more than once.
AI-Accelerated Cost Optimization: The Next Frontier
The teardown relied on proven FinOps practices, but we’re already seeing early wins from AI-driven cost control. PADISO is actively embedding agentic AI into cloud operations for several clients.
Using AI for spend forecasting
We trained a time-series model on 18 months of detailed billing data using Claude Opus 4.8 to predict cost anomalies 48 hours before they occur. The model ingests tags, CloudWatch metrics, and usage patterns to flag upcoming spikes — like a pending HPC workload that would double the compute bill. This predictive layer gives finance teams acting time, not reactive panic.
Current models like Sonnet 4.6 and Haiku 4.5 also prove useful for generating natural-language summaries of cost and usage reports, making FinOps accessible to the board. Competitors like GPT-5.6 Sol or open-weight alternatives can be used, but our preference leans on Claude’s reasoning depth for complex multi-account analyses.
Intelligent resource scheduling
We’re prototyping an AI agent that learns workload patterns and automatically adjusts instance types, spot allocations, and scaling policies across EKS clusters. This goes beyond static rules; the agent can make cost-speed trade-offs in real time, much like a human operator but with millisecond reaction. Early tests in a Los Angeles media streaming environment showed a further 12% cost reduction on top of the base 40% savings, with zero mid-stream interruptions.
The Nops.io article on SaaS FinOps highlights that intelligent capacity management is the coming disruption. We’re already building it into our platform engineering engagements.
Summary & Your Next Steps
The 40% reduction wasn’t magic — it was disciplined execution across five levers:
- Visibility — Tag everything, set budgets and alerts.
- Right-sizing — Match instance types and sizes to workload profiles, optimize containers and storage.
- Architecture — Buy Reserved Instances and Savings Plans for predictable loads, eliminate idle resources, right-size databases.
- Data transfer — Minimize cross-AZ and NAT gateway charges, consolidate VPC peering.
- Continuous optimization — Build a FinOps culture, automate anomaly detection and non-production scheduling.
Here’s how to start: pull your last three AWS invoices, run a 30-day utilization analysis, and identify your top three cost line-items. For each, estimate the savings from the steps above. You’ll likely find the same pattern: 15–20% wasted is common, and 40% is achievable with methodical effort.
If you’ve already done the basics, consider the AI edge. Our platform development teams across the United States and Australia are prototyping agentic cost optimizers that could deliver the next tranche of savings.
How PADISO Helps B2B SaaS Leaders Own Their Cloud Economics
PADISO is a founder-led venture studio and AI transformation firm. We work with mid-market SaaS, PE portfolio companies, and growth-stage startups to drive measurable cloud ROI. Our CTO-as-a-Service engagements give you senior technical leadership to replicate the teardown above — from Atlanta to Sydney.
We also offer platform engineering in Ottawa, Canberra, and more than a dozen cities, always aligning architecture with cost-efficiency and compliance needs — whether that’s SOC 2 in San Francisco or IRAP in Canberra.
Our case studies show real results across industries. If your cloud bill is growing faster than revenue, or you simply want an outsider’s forensic eye on your AWS spend, we’re ready to talk. Reach out for a zero-obligation cloud cost diagnostic — the same one that found $96,000 per month in savings for the client in this teardown.