How Fi Money Cut Kubernetes Costs by 67% with DevZero
Right-sized heavily oversized workloads based on actual usage patterns
Reduced infrastructure spend through continuous workload optimization
Reduced staging CPU consumption with UAT memory utilization dropping 71%
Automated optimization simplified management across hundreds of nodes
About Fi Money
Every day, millions of young Indians open the Fi app, trusting it with something deeply personal: their salary, their savings, their first investment, their financial future. Fi has become the bank account for a generation that grew up on smartphones and refuses to stand in line at a branch. Serving over 3 million users and counting, their platform processes the kind of real-time financial decisions — loans approved in minutes, investments made in seconds — where downtime isn't an inconvenience, it's a crisis. Keeping that promise at scale while migrating to modern cloud infrastructure meant Fi's engineering team was quietly sitting on a ticking cost problem. They were paying for nearly nine times the compute they actually needed — and they had no idea.
The Challenge: Scaling Fast, Paying for 9x More Than You Need
Fi's infrastructure team, led by founding engineer Prasanna Ranganathan, managed a complex and growing cloud footprint: 80% of their compute spend went to critical backend banking services running on EC2 Auto Scaling Groups, with the remaining spend distributed across three main EKS clusters for production services, CI/CD pipelines, and data workloads.
As Fi continued migrating services from EC2 to Kubernetes, several compounding problems emerged:
- Massive over-provisioning: Engineers were manually guessing resource requests with no data-driven guidance. In some workloads, CPU requests were 88% higher than actual usage, meaning the company was paying for nearly nine times the compute it actually needed.
- Wrong instance types: Memory-heavy workloads were running on CPU-optimized instances with a 1:4 CPU-to-memory ratio, when they should have been on instances with 1:6 or 1:8 ratios. This mismatch further inflated costs.
- No cost visibility: The team lacked cluster-level efficiency metrics, making it impossible to identify where waste was occurring to build a business case for optimization.
- Manual, reactive management: With hundreds of nodes across their Kubernetes clusters and no automation, the infrastructure team was spending valuable time on manual cluster management rather than on strategic engineering work.
The Solution: Automated Kubernetes Optimization; Zero Engineering Overhead
Prasanna evaluated the landscape and chose DevZero based on several key factors:
- Low-risk, read-only start: DevZero's read-only operator allowed Fi to gain immediate visibility into cluster efficiency without making any changes to their infrastructure. This was critical for a regulated fintech environment where stability is non-negotiable.
- Hands-on engineering partnership: DevZero's team worked directly alongside Fi's engineers, troubleshooting issues, tuning policies, and iterating on optimizations in real time.
- Quick proof of value: Within the first week of the read-only operator deployment, Fi gained actionable insights into its cluster's efficiency. Within the first month, measurable savings were already materializing.
- Phased rollout approach: DevZero supported a cluster-by-cluster expansion, allowing Fi to validate results at each stage before broadening the scope of optimization.
DevZero's Platform Capabilities
DevZero's platform addresses Kubernetes cost optimization across three interconnected capabilities:
Workload Autoscaling: Automatically right-sizes CPU and memory requests based on actual usage patterns. Policy-driven optimization with configurable guardrails ensures that changes are safe and predictable, eliminating the manual guesswork that led to 88% over-provisioning.
Node Autoscaling: Intelligently provisions and consolidates nodes based on actual workload demand. Cost-aware instance selection matches workload profiles to the right instance families, integrating with AWS Savings Plans and Spot instances for maximum efficiency.
Cluster Visibility: Real-time cost and utilization dashboards provide multi-cluster management views and resource efficiency metrics. For the first time, Fi's team could see exactly where waste was occurring and quantify the impact of optimizations.
“I used to dread deployment days because we never quite knew if a resource issue was going to cause problems in staging that wouldn't show up until prod. DevZero just removed that anxiety entirely. Things behave the way you expect them to.”
The Results: up to 67% Cost Reduction Before Production Was Even Touched
DevZero's impact on Fi's infrastructure has been significant and measurable. The following results reflect validated metrics from two clusters after workload and node automation were fully deployed. These percentage reductions represent efficiency gains achievable in Kubernetes environments with similar over-provisioning patterns.
Staging: * 47% Cost Reduction * 41% CPU reduction * 42% memory reduction
UAT: * 67% Cost Reduction * 61% CPU reduction * 71% memory reduction
“Achieving 67% cost reductions on non-production clusters alone, before we'd even touched production, made the ROI case impossible to argue with.”
Additional Efficiency Gains
DevZero also identified a 50% node reduction opportunity across Fi's broader infrastructure, representing a significant additional efficiency gain as optimizations extend to production workloads.
- CPU minutes reduced by 60% in the staging cluster, reflecting dramatically more efficient resource allocation.
- Underutilization dropped from 93% to 82% in the staging cluster, with further improvements ongoing as policies are refined.
- Instance type optimization shifted workloads from CPU-optimized to memory-optimized instances, better matching actual usage profiles and reducing per-unit cost.
- Engineering time reclaimed: With automated optimization in place, Fi's founding engineering team shifted from manual infrastructure tuning to higher-value product development.
Looking Ahead
DevZero gave Fi's lean infrastructure team the confidence to continue their strategic migration from EC2 to Kubernetes, knowing costs would be automatically optimized as new workloads come online.
If 67% cost reductions are achievable on non-production clusters, the potential impact on production environments — where workloads are larger, more dynamic, and more prone to over-provisioning — is substantially greater. The identified 50% node reduction opportunity across Fi's entire infrastructure underscores how much untapped efficiency remains.
Fi continues to expand its use of DevZero across additional clusters and workload types. Key areas of ongoing and future optimization include:
“DevZero shifted our capacity planning from guesswork to evidence-based scaling — our engineers can focus on building the banking platform instead of managing infrastructure.”
- Full production cluster optimization with workload and node autoscaling
- Realizing the identified 50% node reduction across their entire infrastructure
- Deeper integration with AWS Savings Plans strategy
- Continued migration of EC2 workloads to optimized EKS with confidence in cost control