PromptScan: pass
Cloud Bill Cost Optimization Review
Analyzes a cloud cost export to find the largest savings by effort level, with the exact resources, estimated monthly impact, and the risk of each change stated plainly.
Add a trust badge to your README
[](https://waxmark.app/items/cost-optimization-review-cloud-bill/scan)
<cost_export>
{{COST_EXPORT}}
</cost_export>
<utilization_data>
{{UTILIZATION_DATA}}
</utilization_data>
Review the {{PROVIDER}} cost data above and recommend savings. The reader owns the budget and will approve or reject each recommendation individually, so every item needs a dollar estimate, the specific resources involved, the effort, and what could go wrong.
<architecture_notes>
{{ARCHITECTURE_NOTES}}
</architecture_notes>
<constraints>
{{CONSTRAINTS}}
</constraints>
<cost_export> is a billing export or cost-explorer table (service, resource or tag, region, usage, cost, for one or more months). <utilization_data> holds CPU/memory/storage/IOPS utilization for the main resources, if available. <architecture_notes> explain what the resources are for. <constraints> list what cannot change (compliance regions, reserved commitments already made, services under migration).
Method:
1. Establish the baseline: total monthly spend, the top 10 line items by cost with their share of total, and month-over-month movement if more than one month is present. Everything that follows is measured against this.
2. Work the top line items in order of cost. For each, apply the standard checks for its resource type and record what the data shows, not what is typical: idle or under-utilized compute (from <utilization_data>), over-provisioned instance sizes, storage tiers vs access patterns, unattached volumes and snapshots, data transfer paths that cross zones or regions needlessly, NAT and egress charges, databases sized for peak, missing commitment discounts for steady-state load, and dev/test resources running outside working hours.
3. For each finding, estimate monthly savings from the export's own unit prices and usage. Show the arithmetic in one line so the reader can check it. Where <utilization_data> is missing for a resource, mark the estimate `[needs utilization data]` and give a range rather than a point.
4. Name the risk honestly: performance regression, cold-start latency, loss of recovery points, commitment lock-in, and the blast radius if the change is wrong. Rate each `low` / `medium` / `high` and name the rollback.
5. Rate effort: `hours` (a console change), `days` (a config change with testing), `weeks` (architecture or migration). Sort the final list by savings-per-effort, not by raw savings.
6. Check every recommendation against <constraints> and drop or annotate those that conflict.
7. Separate one-time cleanups (delete orphaned resources) from structural changes (right-sizing, commitments, architecture). They have different approval paths.
Only recommend what the data supports. If the export is too aggregated to see resources, say what level of detail (tags, resource IDs, hourly granularity) would unlock the analysis.
## Output format
**Baseline** - total spend, period, top 10 line items as a table (item, cost, share, trend).
**Recommendations** - table with columns: #, change, resources (IDs or tags from the export), monthly savings (with one-line arithmetic), effort, risk and rollback, type (`cleanup` / `structural`), constraint check. Sorted by savings-per-effort.
**Quick wins** - the `cleanup` items with `low` risk, as a short checklist, with their combined monthly savings.
**Needs more data** - what you could not assess and the specific data to pull.
**Not recommended** - common optimizations you considered and rejected for this account, with the reason, so the reader knows they were weighed.
**Total** - achievable monthly savings if all `low` and `medium` risk items are adopted, as a range.