| name | finops-budget |
| description | Controls cloud and SaaS spend — cost visibility, budgets, alerts, and right-sizing. Use when bills spike, planning infra for a feature, or setting cost guardrails before scale. |
FinOps Budget
Performance without cost awareness burns runway. This skill makes spend visible, budgeted, and
actionable before a surprise invoice — not after.
Pairs with [[observability]] for metrics, [[perf-budget]] for efficiency, [[launch-readiness]] before
scale events, and [[data-modeling]] when storage drives cost.
When to Use
- Monthly cloud bill review or spike investigation
- New feature with infra impact (compute, storage, egress, LLM tokens)
- Setting budgets/alerts for prod
- Right-sizing over-provisioned resources
- Choosing between managed services by cost
Skip for local-only development with no cloud spend.
Process
1. Baseline — know current spend
- Top 5 cost drivers (service/SKU)
- Cost per environment (prod vs staging)
- Trend: last 3 months MoM
- Untagged or "unknown" spend percentage
No baseline = no improvement.
2. Attribute cost to product
| Tag / dimension | Use |
|---|
service / team | Ownership |
environment | prod vs non-prod |
feature | New launch tracking |
Target: >90% tagged spend.
3. Set budgets and alerts
- Monthly budget per env with 50/80/100% alert thresholds
- Anomaly detection on daily spend
- LLM/API: cost per 1k requests if applicable
Document who gets paged at 100%.
4. Optimize — measure before cutting
Priority order:
- Waste — idle resources, orphaned disks, oversized dev clusters
- Right-size — CPU/memory after 2 weeks of metrics
- Storage — lifecycle policies, compression, archive cold data
- Egress — CDN, regional placement, cache ([[caching-strategy]])
- Commitments — reserved/savings plans only after stable baseline
5. Feature cost estimate (before build)
For new features document:
- Compute delta (requests × duration × price)
- Storage delta (GB × retention)
- Egress delta
- Third-party API (LLM tokens, maps, etc.)
- Monthly estimate at 1x, 10x, 100x users
Hand to product-manager if estimate breaks business model.
6. Review cadence
- Weekly: anomaly check
- Monthly: budget vs actual, top deltas
- Quarterly: architecture cost review
Common Rationalizations
- "We're too small for FinOps" — Small teams feel bill spikes hardest.
- "Staging can mirror prod size" — Staging at 25% spec saves 75% there.
- "Reserved instances now" — Wait for 30-day stable usage pattern.
Red Flags
- No billing alerts configured
- Prod and staging indistinguishable in cost reports
- LLM features without token/cost logging
-
20% untagged spend
- Cost discussed only at fundraise time
Verification
