Every cloud cost review we've run this year has found the same pattern: spend growing faster than actual usage. Some of that is genuinely new workloads — more teams running inference workloads, more environments spun up for experimentation. But a large share of it, consistently twenty to thirty percent in the accounts we audit, is waste that's been quietly accumulating because nobody's job is to clean it up.
This is the playbook we actually run with clients — not a checklist of generic best practices, but the specific places we've found real money, in order of how much they usually recover.
Before optimising anything, find what's simply not needed. This is almost always the biggest single win and the easiest to act on.
On a mid-sized account, this pass alone typically recovers five to ten percent of monthly spend, with essentially zero risk — you're removing things nobody is using.
Reserved instances and savings plans are a common next step, and a common mistake if you commit before right-sizing. We've seen accounts locked into three-year reserved capacity for instance sizes that were oversized on day one — the commitment just makes the waste permanent instead of removing it.
Pull actual CPU, memory and network utilisation over the last 30–60 days for every instance class. Anything consistently running under 40% utilisation is a right-sizing candidate before it's a commitment candidate.
Not every workload should be billed the same way. This is where most of the remaining savings sit, once you've cleaned up waste and right-sized:
Most teams set up auto-scaling to handle traffic spikes and stop there — which means the floor of the scaling group is still sized for a worst case that happens a few hours a month. Revisit the minimum instance count with the same scrutiny you gave the maximum. A scaling group that never drops below its peak-hour size isn't actually auto-scaling for cost, only for load.
Compute gets all the attention in cost reviews; data transfer quietly adds up, especially cross-AZ and cross-region transfer between services that were placed without thinking about network topology. If two services talk to each other constantly and are in different availability zones for no architectural reason, that's a recurring cost with an easy fix.
None of the above sticks without accountability. Tag every resource by team, environment and project, and get a monthly cost report in front of the engineering leads who actually control what gets provisioned — not just finance. Waste reappears within a quarter in accounts where cost visibility lives only in a finance dashboard nobody on the engineering team ever opens.
Run this cycle once a quarter and cost stays proportional to actual usage instead of quietly drifting upward. Most teams don't need a bigger cloud budget — they need someone to run this playbook.
This is what our cloud solutions work covers day to day. Tell us where you're stuck and we'll give you an honest read on it.
