The Practices That Actually Work: Cloud Cost Optimisation Without the Mystery

Most cloud bills are not a mystery. They are the predictable result of decisions made years ago that nobody has got around to revisiting — a VM left running after a project wrapped up, reserved instances tied to decommissioned workloads, development environments humming through the weekend with no schedule in place.

The tools to fix this have existed for years. The problem is rarely capability. It is discipline, visibility, and the organisational will to act on what you already know.

Start with visibility, not action

The most common mistake is reaching for the controls before you understand the problem. Proper optimisation starts with a clear picture of where money is actually going — by service, team, environment, and workload — across every provider you use. For multi-cloud teams, pulling AWS, Azure, and GCP billing into a single coherent view requires some work, but it pays off quickly. Without this foundation, optimisation is guesswork. With it, priorities become obvious.

Tagging is the unglamorous backbone

Without a consistent, enforced tagging strategy, you cannot allocate costs to teams or projects with any confidence. Tags applied to 60% of resources are almost as useless as no tags at all, because the remaining 40% is where the surprises hide. The solution is to build tagging into provisioning itself: IaC templates that require tags as mandatory variables, pipelines that validate completeness before deployment. If you have existing tagging debt, prioritise by spend concentration, tag the 20% of resources driving 80% of your bill first.

Right-sizing is the biggest lever most teams pull too gently

The pattern is always the same: a team provisions a large instance, the workload never materialises at expected scale, and the oversized resource runs at 15% CPU utilisation indefinitely. Look at CPU and memory utilisation over at least 30 days. Resources consistently below 20–30% are candidates for downsizing. Done carefully and systematically, right-sizing typically delivers a 20–30% reduction in compute spend.

Commitment discounts: the savings nobody argues about

Reserved Instances, Azure Reserved VM Instances, and GCP Committed Use Discounts all offer 30–60% reductions against on-demand pricing for stable, predictable workloads. The key is committing to a baseline, not a peak, and starting with one-year terms until you have a clear infrastructure roadmap. If more than half your compute spend is running on on-demand pricing for long-running workloads, there is almost certainly money on the table.

Waste elimination is ongoing housekeeping

Orphaned storage, idle load balancers, and development environments running 24/7 accumulate gradually and are nobody's specific responsibility. One engineering team found over £40,000 in annual spend tied to resources from a product feature removed eighteen months earlier. The fix is a regular waste review, monthly or quarterly, with clear owner accountability and a formal decommissioning workflow that includes cost cleanup as an explicit step.

FinOps is a cultural practice, not a project

The organisations that sustain cost discipline over time have made it an operational habit. Engineering teams understand that cost is a concern for engineers. Budget alerts reach engineers directly. Cost reviews are part of sprint retrospectives. And someone explicitly owns the number, with both the authority and the mandate to act.

The cultural shift takes time, but it compounds. Teams that have genuinely internalised cost awareness make better architectural decisions and stop creating waste in the first place.

The cloud is not inherently expensive. It is expensive when it is unmanaged. Managing it well is mostly a matter of deciding to do so.

If cloud costs are a concern and the path forward is unclear, feel free to connect via LinkedIn or reach out through junkshon.com _.