Cloud Cost Review — Sample
A worked example of a monthly cloud cost review — where the money actually goes, the savings that need no architecture, and the difference between waste and a workload in the wrong place.
Markdown. No sign-up, no email.
This is an illustrative example. The organisation, workloads and figures are invented to show the shape of a review that produces decisions. Copy the structure; your numbers will be different.
The review separates three things that get muddled in every cost conversation: waste (spend for nothing), inefficiency (spend for more than needed), and placement (a workload paying rental rates for something it uses constantly). They have different fixes and very different effort.
Cloud cost review — July 2026#
| Period | July 2026, against a 6-month trend |
| Scope | All accounts, one provider |
| Prepared by | Platform team |
| Reviewed with | Engineering leads and the finance business partner |
1. Where the money goes#
| Line item | July | 6-month trend | Share |
|---|---|---|---|
| Compute — production | $41,200 | flat | 38% |
| Compute — non-production | $18,600 | rising | 17% |
| Managed database | $16,900 | rising | 16% |
| Data transfer out | $12,400 | rising sharply | 11% |
| Storage and snapshots | $9,100 | rising | 8% |
| NAT gateway | $5,300 | flat | 5% |
| Everything else | $4,900 | flat | 5% |
| Total | $108,400 | +19% over 6 months |
Volume served grew 6% over the same period. Spend grew three times faster than the thing it pays for, which is the sentence this review exists to produce.
2. Waste — spend returning nothing#
| Item | Monthly | Effort |
|---|---|---|
| Non-production running nights and weekends | $11,400 | A scheduler. One day. |
| Unattached storage volumes (61 of them) | $2,850 | Delete. One afternoon. |
| Snapshots with no retention policy, oldest from 2023 | $2,100 | Policy + one cleanup |
| Load balancers with no traffic for 90 days | $740 | Delete after owner confirms |
| Idle IP addresses | $310 | Delete |
| Total | $17,400 | ~2 days of work |
Sixteen per cent of the bill, and none of it requires an architecture change. The non-production figure alone is the largest single saving available, and it is a scheduled task.
The 61 unattached volumes are worth a note. Nobody deleted them because nobody could establish whose they were — none carry an owner tag. That is why the tagging recommendation below is not administrative housekeeping.
3. Inefficiency — paying for more than is used#
| Item | Current | Measured need | Monthly saving |
|---|---|---|---|
| Application tier sized from the original estimate | 24 instances | 14 at peak | $6,900 |
| Managed database on the largest tier | — | Peak CPU 22% | $4,100 |
| Cross-zone traffic from a service talking to itself | — | Avoidable | $1,800 |
| Logs retained 12 months, queried back 14 days | — | 30-day policy | $1,200 |
| Total | $14,000 |
Right-sizing needs measurement rather than opinion, and the measurements are already available — this is a week of careful work, not a project.
4. Placement — the workload in the wrong place#
One workload accounts for a disproportionate share and is not waste. It is correctly built, fully used, and paying rental rates for something it uses every hour of every day.
| Batch processing cluster | |
|---|---|
| Monthly cost | $22,700 |
| Utilisation | 78% average, 24 hours a day |
| Load pattern | Steady, predictable, no seasonality |
| Data egress | 41 TB per month |
| Expected lifetime | Years — it is core to the product |
| Owned equivalent | ~$180,000 up front, ~$3,400 monthly running |
On these figures the crossover is under a year. This is the one item in the review that is a decision rather than a task, and it belongs to the owner, not to the platform team. It also carries a portability question: if it moves, it must remain movable.
Note what is not being proposed. Nothing else in the estate has this profile. Variable workloads, new services and anything that might not exist in six months stay where they are.
5. Cost controls#
| Control | Status |
|---|---|
| Budget alerts | Set at 80% and 100%, per account |
| Mandatory tags — owner, environment, cost centre | Not enforced — 34% of resources untagged |
| Non-production off out of hours | Not implemented |
| Retention policies on logs and snapshots | Partial — logs only |
| Named owner for the bill | No |
The two blanks in the middle are the two largest savings in this review. That is not a coincidence: uncontrolled cost is a governance outcome before it is a technical one.
6. Summary#
| Monthly | Effort | |
|---|---|---|
| Waste | $17,400 | 2 days |
| Inefficiency | $14,000 | 1 week |
| Available without an architecture change | $31,400 | 29% of the bill |
| Placement decision (separate) | up to $19,300 | Owner decision + a project |
7. Recommendations#
- Schedule non-production off outside working hours — one day, $11,400 a month.
- Enforce tagging, and make untagged resources subject to deletion after notice. Without this, the cleanup below has to be repeated every quarter.
- Delete the identified orphans — volumes, snapshots, load balancers, addresses.
- Right-size against measured usage, starting with the application tier and the database.
- Name an owner for the bill. Every controlled estate has one; no uncontrolled estate does.
- Take the batch cluster to the owner as a placement decision with these figures. Do not fold it into the technical actions above.
Notes on using this format#
Separate waste, inefficiency and placement. Combined, they produce one intimidating number and no action. Separated, most of the saving turns out to need a day and a scheduler.
Put effort next to every saving. It is what makes the order obvious and stops the review becoming a wish list.
Compare spend growth to the thing it pays for. A bill that grows with the business is not a problem. Growing three times faster is, and neither number means anything alone.