Cloud cost is an architecture problem, not a procurement one
Most cloud cost programmes start in finance, produce a respectable one-off saving through commitment discounts and rightsizing, and then plateau while spend resumes its previous growth rate. The savings were real; they addressed the symptom.
The decisions that actually set the bill
Data transfer topology
Cross-region and cross-AZ transfer charges are invisible in architecture diagrams and highly visible on invoices. A chatty service pair split across availability zones can cost more in transfer than in compute.
Storage class and lifecycle
Logs and artefacts retained indefinitely at hot-storage prices are one of the most common large line items. Lifecycle policies are a ten-minute change that nobody owns.
Managed service selection
Managed services trade cost for operational load. That is often correct — but the trade should be deliberate and revisited, not the result of whatever was easiest during a proof of concept.
Making cost visible where decisions are made
Tag everything by team and service, and put the resulting figures in front of engineers weekly. Cost stops being an abstraction the moment an engineer can see that their service costs more than the rest of the platform combined.
The unglamorous wins
- Non-production environments that shut down outside working hours.
- Autoscaling that actually scales down, verified rather than assumed.
- Orphaned volumes, snapshots and load balancers swept automatically.