41% Cloud Cost Reduction in 90 Days : How a SaaS Scale-Up Finally Got Control of Its Cloud Bill

The company had done everything right on the product side. ARR had grown more than four times over in 18 months. The engineering team was shipping. Customers were renewing. By every growth metric, things were working.

But the cloud bill told a different story.

Infrastructure spend had grown more than twice as fast as revenue across the same period — a gap that looked manageable in month three and catastrophic by month eighteen. By the time the CFO escalated it, cloud was consuming over 9% of ARR. For a company approaching its Series B, that number was a dealbreaker. Investors don’t fund runaway cost structures.

The engineering team wasn’t being reckless. They were being cautious in the wrong direction — defaulting to larger instance sizes to avoid performance incidents, leaving non-production environments running around the clock because spinning them back up felt risky, and never cleaning up resources from deprecated services because nobody officially owned that task. Over time, those individual decisions compounded into an invisible drag on the business.

Three internal efforts to fix it had already failed — not because the engineers lacked competence, but because the problem wasn’t a technical one. It was an organisational one. Finance couldn’t see where money was going. Engineering didn’t think in cost terms. Nobody owned the gap between them. And with spend split across AWS and Azure with no unified view, even diagnosing the problem clearly was impossible.

The company needed more than a cost-cutting exercise. It needed a new way of thinking about cloud — one that could survive the next growth phase without collapsing again.

The engagement started with a single rule: no optimisation decisions before the data was clean. Jumping straight to rightsizing or reserved instances — the instinct most teams have — locks in the wrong baseline. The first job was visibility.

Phase 1 — Visibility (Weeks 1–2): A full cloud estate audit was conducted across both AWS and Azure. Using AWS Cost Explorer, Azure Cost Management, and Apptio Cloudability as a unified lens, 100% of spend was mapped to teams, services, and environments for the first time. A mandatory tagging policy — Environment, Team, Product, CostCentre — was enforced at the infrastructure layer via AWS Config rules and Azure Policy. Within ten days, the attribution gap that had made the problem invisible was closed.

Phase 2 — Quick Wins (Weeks 3–5): With clean data, the waste became undeniable. Rightsizing analysis identified over sixty EC2 and Azure VM instances running below 20% CPU and memory utilisation — not because the workloads were light, but because the instances had never been sized to the actual workload. Non-production environments were placed on automated scheduling, shut down every night and weekend via AWS Instance Scheduler and Azure Automation Runbooks. Orphaned resources — EBS volumes, unused Elastic IPs, forgotten load balancers from deprecated services — were reviewed with engineering leads and terminated. The engineering team had been hesitant to touch any of this. Seeing the actual utilisation numbers made the decisions straightforward.

Phase 3 — Commitment Optimisation (Weeks 6–9): Commitment purchases are only safe after a stable, tagged baseline is established. After six weeks of clean data, Reserved Instance and Savings Plan purchases were modelled across a one-year horizon. AWS Compute Savings Plans were applied to the majority of steady-state EC2 workloads, delivering over 40% savings versus on-demand on covered resources. Azure Reserved VM Instances were applied to the ML training cluster — a workload that was predictable but had been paying on-demand rates the entire time.

Phase 4 — Culture and Governance (Weeks 10–12): The goal was never to hand back a lower bill. It was to hand back a team that could never lose control of it again. A bi-weekly Cloud Cost Review was established between Finance, Engineering, and Product. Real-time dashboards were embedded directly in the internal developer portal. Anomaly detection was configured through AWS Budgets and Azure Cost Alerts. A Cloud Cost Runbook, covering rightsizing decision criteria, tagging standards, and commitment review cadence — was written so that the process would survive team changes.

• Unified multi-cloud cost visibility across AWS and Azure : a single dashboard, no more siloed bills

• Mandatory tagging enforced at infrastructure layer : 98% attribution coverage achieved

• Non-production environments automated off every night and weekend : idle compute eliminated

• Over 60 instances rightsized from oversized M5 and D-series to right-fit T3 and B-series

• Savings Plans and Reserved Instances applied to the majority of stable workloads

• FinOps operating model installed: bi-weekly cadence, anomaly alerting, written runbook