How a FinTech Unicorn Reduced Cloud Spend by 42%

The Challenge

NovaPay (name changed for confidentiality) is a fast-growing fintech unicorn processing over $12 billion in annual transactions. With 800 engineers across three continents and a multi-cloud infrastructure spanning AWS, GCP, and Azure, their cloud bill had grown to $5.7 million per month -- a 340% increase over two years -- with no clear understanding of where the money was going or who was responsible for it.

Growing Complexity

NovaPay's infrastructure had grown organically through acquisitions and rapid product development. Each team chose their preferred cloud provider, leading to a fragmented landscape:

• AWS: Core payment processing (60% of spend) -- EC2, RDS, Lambda, S3
• GCP: Data analytics and ML workloads (25% of spend) -- BigQuery, Vertex AI, GKE
• Azure: Enterprise integrations and compliance (15% of spend) -- AKS, Cosmos DB, Key Vault

Three different billing systems, three different tagging conventions (or lack thereof), and three different teams managing costs independently. The CFO was frustrated: monthly cloud spend reports arrived 3 weeks late, contained different metrics depending on the provider, and could not answer basic questions like "How much does our fraud detection service cost across all providers?"

Lack of Visibility

The core problems were clear:

1. No unified view: Cost data existed in three separate systems with incompatible formats
2. No cost attribution: Only 34% of resources were properly tagged. The remaining 66% sat in a catch-all bucket labeled "unallocated"
3. No accountability: Without attribution, no team owned their cloud spending, and optimization was nobody's priority
4. No anomaly detection: A $47,000 spike from a misconfigured auto-scaling group went unnoticed for 11 days

"We were spending $5.7 million a month on cloud infrastructure and couldn't tell our board which products were profitable after infrastructure costs. That's not a billing problem -- it's a business intelligence failure." -- VP of Engineering, NovaPay

The Solution

NovaPay selected CloudAct.ai for its ability to unify multi-cloud cost data into a single FOCUS 1.3-normalized view, combined with its organizational hierarchy model that maps cloud spending to business structure.

The key differentiators that drove their decision:

• Multi-cloud normalization: All three providers' billing data converted to a unified format automatically
• Hierarchy-based attribution: Map resources to departments, business units, and teams using CloudAct.ai's four-level hierarchy
• GenAI cost tracking: NovaPay's ML team was spending $180K/month on Vertex AI and OpenAI with no visibility into per-model or per-pipeline costs
• Real-time alerting: Budget-based alerts to catch anomalies before they become expensive incidents
• Multi-currency support: NovaPay operates in 14 countries -- costs needed to be viewable in USD, EUR, GBP, and INR

Implementation

Week 1: Unified Visibility

Implementation started on a Monday morning. By Friday, NovaPay had full multi-cloud visibility.

Day 1-2: Provider Integration

• Connected AWS Cost and Usage Reports via cross-account IAM role
• Connected GCP billing export (already flowing to BigQuery)
• Connected Azure Cost Management API via service principal
• Set up OpenAI and Vertex AI GenAI cost tracking

Day 3-4: Hierarchy and Attribution

• Defined organizational hierarchy: 4 departments, 12 business units, 38 teams
• Mapped existing tags to hierarchy nodes where possible (34% coverage)
• Identified the 66% unallocated resources and assigned remediation owners

Day 5: First Insights

• Unified dashboard showed total multi-cloud spend for the first time
• Immediately identified $340K/month in orphaned resources across all three providers
• GenAI dashboard revealed that 40% of Vertex AI spend was on a deprecated model variant that was 3x more expensive than the current version

Week 2: Targeted Optimization

Quick wins implemented in the second week:

1. Orphaned resource cleanup: Terminated 127 idle resources identified in the first visibility scan ($340K/month savings)
2. GenAI model migration: Migrated fraud detection ML pipeline from deprecated Vertex AI model to current version ($72K/month savings)
3. Dev environment scheduling: Implemented automatic shutdown of development and staging environments outside business hours ($156K/month savings)
4. Right-sizing Phase 1: Downsized 43 over-provisioned RDS and Cloud SQL instances that were running at less than 15% CPU utilization ($89K/month savings)

Results

Over the first 90 days, NovaPay achieved transformational results:

Metric	Before	After (90 days)	Improvement
Monthly cloud spend	$5.7M	$3.3M	-42% ($2.4M/year)
Tag compliance	34%	91%	+57 percentage points
Cost report latency	3 weeks	Real-time	Eliminated lag
Anomaly detection	11 days average	15 minutes	99.9% faster
Cost per transaction	$0.047	$0.027	-43%
GenAI cost visibility	None	Per-model, per-pipeline	Full attribution

The 42% cost reduction broke down into:

• Orphaned resources: $340K/month (6% of total)
• Right-sizing: $520K/month (9.1%)
• Commitment discounts: $890K/month (15.6%) -- applied after 60 days of stable baseline data
• Environment scheduling: $156K/month (2.7%)
• GenAI optimization: $180K/month (3.2%)
• Storage lifecycle policies: $314K/month (5.5%)

Key Takeaways

NovaPay's experience highlights several universal lessons for multi-cloud cost optimization:

1. Visibility must come first. You cannot optimize what you cannot see. Unifying multi-cloud data into a single normalized view was the prerequisite for every subsequent optimization.
2. Quick wins build momentum. Orphaned resource cleanup and environment scheduling delivered immediate savings that built executive confidence and team buy-in for deeper optimizations.
3. GenAI costs need the same rigor as cloud costs. NovaPay's $180K/month GenAI spend was completely invisible before CloudAct.ai. Treating AI costs as a first-class cost category prevented waste from growing further.
4. Hierarchy drives accountability. Mapping cloud resources to business structure turned cloud costs from an abstract infrastructure problem into a team-level responsibility.
5. Speed matters. The 14-day time-to-value with CloudAct.ai meant NovaPay started saving money in the same billing cycle they deployed the tool. Every week of delayed optimization is wasted money.

"CloudAct.ai gave us something we never had: a single source of truth for our multi-cloud costs. Within two weeks, we found $2.4M in annual savings. Within 90 days, we had a FinOps practice that our CFO actually trusts." -- VP of Engineering, NovaPay