cost-optimization

Files (3)

• SKILL.md
• _claude-frontmatter-extras.yaml · references/
• cloud-audit-checklist.md · references/

SKILL.md (first lines)

# Cost Optimization

Audit cloud, SaaS, and infrastructure spend. Cut what's not earning its keep. Rightsize what's oversized. Negotiate what's negotiable. Without breaking what works.

---

## When to use

- Quarterly or annual cost review
- Finance flags rising spend
- Vendor contract renewal coming up
- Budget cut required
- New leadership wants the numbers
- Migrating between providers (cost is part of the case)
- Audit before scaling significantly (catch waste before it scales)

## When NOT to use

- Active incident response (use `incident-response`)
- Performance issues that happen to involve infrastructure (use `performance-optimization`)
- Vendor evaluation for a new purchase (use `vendor-evaluation`)
- Personnel or org costs (out of scope for this skill)

---

## Required inputs

- Current cost (monthly, ideally for the last 12 months)
- Cost broken down by service or vendor
- Inventory of cloud resources (instances, databases, storage, etc.)
- Inventory of SaaS subscriptions
- Owners per cost line (who decided to spend this, who uses it)
- Constraints (compliance, performance, contract terms)

---

## The framework: 5 levers

Every cost optimization opportunity falls into one of these levers.

### Lever 1: Eliminate

Stop paying for things that aren't used.

- Idle resources (instances, databases, environments running but unused)
- Subscriptions where no one logs in
- Duplicate tools (multiple tools doing the same job)
- Old projects still incurring cost
- Test environments that should have been torn down
- Forgotten domains, backups, snapshots, logs

This is usually the largest opportunity in the first audit. Often 10-30% of spend.

### Lever 2: Rightsize

Pay for what you actually use, not what you provisioned for the worst case three years ago.

- Oversized instances (CPU and memory utilization low)
- Over-provisioned databases (storage and throughput far above usage)
- Over-purchased SaaS seats
- Premium plans where standard would suffice
- High-availability setups for non-critical systems

Rightsizing requires real usage data, not theoretical needs.

### Lever 3: Restructure

Use cheaper structures for the same workload.

- Reserved or committed-use pricing (1-3 year commitments at 30-70% discount)
- Spot or preemptible instances for fault-tolerant work
- Cold storage for data accessed rarely
- Tiered storage (hot/warm/cold) by access pattern
- CDN caching to reduce origin load
- Compression and deduplication
- Serverless for spiky workloads
- Reserved instances for steady workloads

The right structure depends on the access pattern. Mismatch costs money.

### Lever 4: Negotiate

Pay less for the same thing.

- Annual contracts at lower rates than monthly
- Volume discounts at higher tiers
- Multi-year commitments for predictable workloads
- Bundle deals (consolidating services with one vendor)
- Renewal negotiation (vendors expect you to ask)
- RFP / competitive bid (using alternatives as leverage)

Most enterprise vendors negotiate. Most SaaS vendors don't, except at higher tiers. Consumer-tier services usually don't.

### Lever 5: Reframe

Change the question.

- Build vs buy: maybe in-house is cheaper at scale
- Buy vs build: maybe outsourcing is cheaper at small scale
- Different architecture (e.g., monolith vs microservices) has different cost profiles
- Different audience (do all customers need the same tier?)
- Different stack (open source vs commercial)

Reframe is the longest-lead lever. Worth thinking about even if not actionable now.

---

## Workflow

### Step 1: Pull the spend data

Get monthly costs by service, vendor, and (where possible) team or project.

For cloud (AWS, GCP, Azure): the billing console and cost-explorer tools.
For SaaS: each vendor's billing portal, plus an SaaS-management tool if available.
For everything else: bank statements and accounting export.

12 months minimum. Trends matter as much as absolute numbers.

From rampstackco/claude-skills, MIT.

More from rampstackco/claude-skills