Cutting a scaling startup's cloud bill by 60%

// the challenge

A startup was scaling fast on Kubernetes and paying for far more compute than it used. Nodes were sized for peak and ran half-empty the rest of the time, staging environments stayed on around the clock for work that happened during business hours, and old volumes and snapshots were never cleaned up. None of it was visible day to day, so the waste compounded quietly — and the team had no spare time to chase it.

// our approach

• Audited real usage across every cluster to separate genuine demand from padding.
• Right-sized node pools and set up autoscaling so capacity tracks actual load.
• Moved fault-tolerant workloads onto spot capacity and committed steady baseline load to reserved capacity, cutting the rate paid per unit of compute.
• Added cost dashboards and alerts so spend stays visible to the team and waste can't silently creep back.
• Built an agent to make optimisation autonomous so the waste doesn't come back after we leave.

// the outcome

Monthly cloud spend fell by around 60% while the team kept shipping daily. Utilization roughly doubled — the same work now runs on far less idle capacity. The team has better visibility into cost and less ongoing maintenance overhead.