Why Cloud Bills Spiral
Cloud cost overruns are almost always a governance problem, not a technical one. Engineers provision infrastructure to solve immediate problems, nobody reviews whether that infrastructure is still needed, and the bill compounds silently. The first audit finding is almost always the same: instances running at 12% average CPU utilisation that were sized for a peak load that happened once, eight months ago.
The bill is high because no one owns it not because the architecture is wrong.
Right-Sizing Compute
Right-sizing is the highest-leverage optimisation in most accounts. Check CloudWatch or Cloud Monitoring metrics for your running instances. If your p99 CPU utilisation over the past 30 days is below 40%, you’re almost certainly over-provisioned. Most teams find they can reduce compute spend by 25–35% with zero application changes. Start with your ten largest compute line items, drop each to the next size tier, monitor for a week, and repeat.
Reserved Capacity Strategy
A 1-year reserved instance on AWS saves roughly 40% over on-demand pricing for the same compute. The right approach: identify your stable baseline compute the instances running 24/7 that you’re confident you’ll need for at least a year purchase reserved capacity for those, and use on-demand or spot for variable workloads on top. Most companies are undercommitted on reservations relative to their stable workload.
Data Transfer Costs
Data transfer is the cloud bill line item that surprises people most, because it’s invisible until it’s enormous. The most common culprit: application servers and databases in different availability zones, causing cross-AZ transfer on every database query. Move them to the same AZ. Use VPC endpoints for S3 and DynamoDB to eliminate NAT Gateway costs. Cache aggressively at the edge to reduce origin egress.
Storage Lifecycle Policies
Implement S3 lifecycle policies on every bucket: transition to Infrequent Access after 30 days, Glacier after 90 days, and delete after your required retention period. Apply the same logic to EBS snapshots, CloudWatch log groups, and ECR container image repositories. Most teams find 20–30% of their storage bill is data that nobody reads anymore logs, build artifacts, and database backups that accumulated silently.
Building Accountability
The technical fixes only stick if someone owns the cloud budget. Assign a FinOps owner even part-time whose responsibility includes a monthly cost review and a target to hit. Implement cost allocation tags so you can break the bill down by service, team, and environment. Set billing alerts that fire before you exceed budget. The teams that sustain cost improvements treat cloud spend as an engineering metric, not a finance line item.
- Right-sizing compute is the highest-leverage starting point check utilisation before anything else
- Reserved instances save ~40% on stable workloads commit to them early
- Cross-AZ data transfer is often invisible until it’s expensive co-locate your services
- S3 lifecycle policies on every bucket, always unmanaged storage compounds fast
- Assign a FinOps owner governance is what makes savings stick long-term