SaaS Development Company in Seattle

Cloud Infrastructure / FinOps

A Seattle-based startup (Seed, $4.5 million raised, founded by two ex-AWS engineers) was building a FinOps SaaS platform — helping mid-market engineering teams (100-1,000 engineers) understand and optimise their AWS and Azure cloud spend. The founders had built a prototype that connected to AWS Cost Explorer and generated basic spend reports. The prototype's problems: data latency (Cost Explorer data is delayed 24-48 hours — customers wanted near-real-time spend visibility), limited recommendations (the system identified obvious waste like idle EC2 instances but missed the complex optimisation opportunities — Reserved Instance/Savings Plan planning, right-sizing, and architecture-level cost improvements — that represent 60% of cloud cost savings), single-cloud only (AWS-only — 40% of target customers use multi-cloud, and even AWS-primary customers increasingly have Azure workloads), no team attribution (engineering leaders needed to understand cost by team, service, and project — but the prototype could only show cost by AWS account and resource), and no workflow (identifying savings is useful but implementing them requires engineering changes — the prototype had no mechanism for creating, assigning, and tracking optimisation tasks). The founders needed to reach $1M ARR within 12 months to raise a Series A.

We rebuilt the platform as a multi-cloud FinOps SaaS with near-real-time cost visibility, ML-driven recommendations, team attribution, and optimisation workflow. (1) Multi-cloud data pipeline: ingesting cost and usage data from AWS (Cost and Usage Reports via S3, real-time via CloudWatch billing metrics, Reserved Instance and Savings Plan utilisation via AWS Cost Explorer API), Azure (Cost Management exports via Storage Account, real-time via Azure Monitor billing metrics, Reserved Instance utilisation via Azure Advisor API), and infrastructure metadata (EC2 instance types, RDS configurations, Lambda invocations, EKS cluster sizes — the context needed to make optimisation recommendations). Data freshness: 4-hour delay (from 24-48 hours) — achieved by combining CUR data (comprehensive but delayed) with CloudWatch billing metrics (near-real-time but less detailed). Processing pipeline: daily full refresh + 4-hourly incremental updates using AWS Step Functions and DynamoDB for the cost data store. (2) ML-driven recommendations: beyond simple "idle instance" detection. Savings Plan optimisation: ML model that analyses 12 months of usage patterns to recommend optimal Savings Plan coverage — balancing commitment level against flexibility, considering seasonal patterns and growth trends. The model runs Monte Carlo simulations of future usage scenarios and recommends the commitment that minimises expected cost across scenarios. Right-sizing: analysing CloudWatch CPU, memory, network, and disk metrics for every compute instance and recommending the smallest instance type that maintains performance — accounting for periodic spikes (a server that's 5% CPU average but 90% during daily batch jobs shouldn't be downsized). Architecture recommendations: identifying patterns that suggest architectural cost improvements — "this Lambda function is invoked 500M times/month at $0.20/1M invocations = $100/month, but the average duration is 15ms — migrating to a container would cost $18/month" — these architecture-level recommendations require understanding both pricing and engineering trade-offs. Anomaly detection: identifying unexpected cost spikes within hours (from days with the old system) — "your us-east-1 RDS spend increased 340% in the last 4 hours, likely caused by the new db.r5.4xlarge instance launched at 14:23 UTC." (3) Team attribution: mapping cloud costs to engineering teams and services. Tag-based attribution: using AWS tags and Azure resource groups to attribute costs to teams, projects, and environments. Tag coverage dashboard: showing which resources are untagged (preventing cost attribution) — with automated tag recommendation based on resource naming patterns and account structure. Service-level attribution: for Kubernetes workloads, attributing costs to individual services based on pod resource requests and actual utilisation — not just the total EKS cluster cost but cost per microservice. Showback reports: automated monthly reports per engineering team showing: total spend, month-over-month change, cost per unit (per customer, per API call, per transaction — connecting cloud cost to business metrics), and available optimisation opportunities specific to that team's resources. (4) Optimisation workflow: turning recommendations into engineering tasks. Each recommendation becomes an actionable ticket with: description (what to change), expected savings (monthly and annual), implementation effort (estimated engineering hours), risk assessment (what could go wrong — e.g., right-sizing risk if the instance occasionally needs more capacity), and implementation instructions (step-by-step guide including Terraform code changes where applicable). Integration with Jira and Linear for task creation — recommendations flow directly into engineering sprint backlogs. Status tracking: approved, in-progress, implemented, validated — with automatic savings validation once the change is deployed (did we actually achieve the predicted savings?). (5) Executive dashboard: for engineering VPs and CFOs. Real-time cloud spend with forecast (current month projected spend based on trajectory, compared to budget), unit economics (cost per customer, cost per transaction, cost per API call — the metrics that matter for SaaS business models), optimisation pipeline (total identified savings, savings in-progress, savings achieved — showing the ROI of the FinOps programme), and benchmarking (anonymous comparison against similar-sized companies — "your compute cost per $1M ARR is $42K, the median for companies your size is $35K — you're spending 20% more than peers").

Customer acquisition: 48 paying customers in 12 months, reaching $1.3M ARR (30% above target). Average contract: $27K ARR. Customer profile: mid-market SaaS companies (100-800 engineers, $5M-$50M annual cloud spend). Average customer savings: 31% reduction in cloud spend within 6 months of implementation. For a customer spending $2M annually, that's $620K in savings — against a $27K annual platform fee, the ROI is 23x. Total customer savings: $18.4 million aggregate savings across all customers in year one. Savings breakdown: Reserved Instance/Savings Plan optimisation (42% of savings), right-sizing (28%), idle resource elimination (18%), and architecture recommendations (12%). Recommendation accuracy: 89% of implemented recommendations achieved within 10% of predicted savings — giving customers confidence to implement changes. Anomaly detection: the platform identified 340 cost anomalies across customers in year one, with average detection time of 3.8 hours (from days). Average anomaly cost: $2,400 — early detection prevented an estimated $816K in cumulative waste. Team attribution: 94% of cloud spend attributed to teams within 30 days of platform deployment (from 0% visibility). Engineering leaders report that cost visibility changed team behaviour — teams that see their costs make more cost-conscious architecture decisions. Optimisation workflow: 1,800 recommendations generated, 1,240 approved, 980 implemented, 840 validated with actual savings. The workflow integration with Jira/Linear was cited by 78% of customers as "essential — without it, recommendations would sit in a dashboard and never be implemented." Series A: the $1.3M ARR, 48 customers, and 31% average savings metrics contributed to a $22 million Series A from a top-tier FinOps-focused investor. Development investment: $340,000 over 14 weeks.