Analyze my cloud spend and give me the changes worth making, in order.
Provider and accounts in scope: {{e.g. AWS, two accounts (prod + staging)}}
Monthly spend and trend: {{e.g. $84k last month, up 11% over three months}}
Top billing lines: {{paste the top 10-15 lines from the cost report: service, usage type, monthly cost, month-over-month change}}
How the main workloads run: {{e.g. the API is steady 24/7 with a 2x daytime peak; the data pipeline runs nightly for 3 hours; staging is idle outside office hours; the training jobs are bursty and restartable}}
What is already committed: {{e.g. a 1-year savings plan covering roughly 40% of compute; no reserved database capacity; none}}
Constraints: {{e.g. no region changes; the database cannot have downtime; the team is two engineers and cannot take on a re-architecture this quarter}}
Target: {{e.g. 20% reduction within 90 days}}
For each top billing line, separate the two kinds of lever:
- **Usage levers**, which change what we run: right-sizing on observed p95 utilization, scheduling non-production off outside hours, deleting idle and orphaned resources (unattached volumes, old snapshots, idle load balancers, unused addresses), storage lifecycle tiering, log retention, and moving restartable jobs to spot or preemptible capacity.
- **Rate levers**, which change what we pay for the same usage: commitments (reserved capacity, savings plans) sized against the steady baseline rather than the peak, and pricing-tier changes.
For every lever, give: the estimated monthly saving as a range, the confidence in that range and what it rests on, the production risk stated concretely (what breaks, and whether it reverses in minutes or in days), the effort in engineer-days, and the one measurement to take before doing it.
Rank everything by savings against risk and effort, then split into three lists:
**This week.** Reversible, no user impact, no approval needed beyond the team. Cleanup and scheduling usually live here.
**This quarter.** Needs a measurement period or a change window. Right-sizing and commitments live here; state the baseline period a commitment should be sized from, and why buying against the current peak is the mistake to avoid.
**Not now.** Levers that exist but fail my constraints. Name the constraint, so the list is honest about what the target would take if that constraint moved.
Close with whether the target is reachable from the first two lists alone, and the guardrail that keeps the savings from creeping back: the budget alert, the tagging policy, or the scheduled audit for orphaned resources.
Do not propose a lever for a line whose usage pattern you cannot see; ask for the pattern instead. Every number you did not read from my report is an estimate and should say so.Tip: Export the cost report grouped by service and usage type rather than by resource; resource-level exports are too long to paste and the levers live at the usage-type level anyway. Account IDs and resource names are not needed for the analysis.
finopscloud-costinfrastructurebudgeting