Most cloud cost advice is a list of dashboards to buy. This is a list of things to change. Cloud spend leaks in four predictable places, and closing them cuts a double-digit percentage off the bill without a migration. Here is where the money goes, how to fix each leak, and a live tool that diagnoses it from your own billing data with nothing leaving your browser.
Video and audio walkthroughs of this topic. Media slots below are ready for embeds.
Cloud spend does not usually balloon from one bad decision. It grows the way a garage fills up: a little at a time, with no one responsible for throwing anything out. An engineer provisions a large instance to be safe, a team spins up a cluster for a project that ends, a service gets tagged for one quarter and never again. Each is individually reasonable. Together they are the reason your bill is 20 to 40 percent higher than the work you are actually doing requires. The fix is a short, repeatable checklist run against your real billing data, plus one owner who closes the loop.
| Leak | Signal | Where it hides | The fix |
|---|---|---|---|
| Idle / over-provisioned | Under ~20% utilization | Big line items: warehouses, clusters, oversized instances | Rightsize the expensive ones first; auto-stop non-prod |
| Untagged spend | No owner, environment, or cost-center tag | Anything spun up fast and never labeled | Enforce a minimal tag policy; treat untagged as a defect |
| Concentration risk | One service is a large share of the bill | Your top spender | Put the deepest effort where the money is, not evenly |
| Budget drift | Spend exceeds plan with no alert | Between the change and the invoice | Budget per team, alert at 85%, escalate on overrun |
Any resource running well below its capacity is money spent on air. The rule of thumb: anything under roughly 20 percent utilization is a rightsizing or shutdown candidate. Look hardest at the expensive line items, an over-provisioned data warehouse or a barely-used Kubernetes cluster dwarfs a hundred idle small instances. Rightsize the big ones first, set auto-stop on non-production, and delete what no one can explain.
Untagged spend is spend with no owner, which means no one is incentivized to cut it. Enforce a minimal tag policy (owner, environment, cost-center) and treat untagged resources as a defect, not a nuisance. The goal is not perfect tagging, it is that every significant line item traces to a team that will answer for it.
Untagged spend is unowned spend, and unowned spend is uncut spend. The tag policy is not bureaucracy, it is the list of who you can send the bill to.
Find untagged line itemsWhen a single service is a large share of your total spend, it is both your biggest optimization opportunity and your biggest budget risk: a small percentage change there swamps everything else. Know your top spender's share of the total, and put your deepest optimization effort where the money is, not evenly across a hundred small services.
See spend by serviceSet a monthly budget per team or workload, alert at 85 percent, and escalate on overrun. The point is not to block spend, it is to make a spike visible the day it starts instead of on the invoice a month later.
Check budget varianceThe share of your bill in over-provisioned or untagged resources. Your optimization headroom.
Dollars with no owner. Unowned spend is uncut spend.
The largest service's share of total. Rising concentration means rising budget risk.
Actual versus budget per team. If you measure one thing monthly, measure this.
Find the waste in your own bill in an afternoon.
Diagnose my bill