FinOps
You know what you pay per transaction. Not per server.
Cloud cost optimization as a continuous process. Tagging, visibility, right-sizing, reserved instances. Typically 30-50% savings.
Why cloud costs more than it should¶
Most companies pay 30-50% more for cloud than they need to:
- Oversized VMs — 4 CPU and 16 GB RAM for a service that needs 0.5 CPU
- Forgotten resources — Test environments from last year, nobody turned them off
- No reserved instances — Paying on-demand for stable workloads
- Storage without tiering — Logs from 2022 on premium storage
- No autoscaling — Peak capacity running 24/7
Cost Visibility¶
You can’t optimize what you can’t see. The first step is tagging:
- Team — Which team owns the resource
- Environment — dev/staging/prod
- Project — Which project it belongs to
- Cost center — Financial allocation
Showback dashboard per team. When a team sees their dev environment costs 50K/month and runs 24/7, they’ll come up with autoscaling ideas themselves.
Optimization Techniques¶
Quick wins (week 1-2): - Right-sizing oversized VMs (CPU/memory utilization analysis over 30 days) - Shutting down unused resources (unattached disks, idle LBs, forgotten VMs) - Dev/staging scale-to-zero outside business hours
Medium-term (month 1-3): - Reserved Instances / Savings Plans (30-60% discount for commitment) - Spot/preemptible instances for non-critical workloads (60-90% discount) - Storage tiering (hot → cool → archive) - Autoscaling for variable workloads
Long-term (3-6 months): - Architectural changes (serverless for event-driven workloads) - Multi-region optimization - Data pipeline optimization (reduce egress) - Unit economics tracking
Unit Economics¶
The most important FinOps metric: cost per business unit.
- Cost per transaction
- Cost per active user
- Cost per API call
- Cost per GB processed
When you know unit cost, you can: optimize meaningfully, forecast accurately, communicate with business. “We cost 0.003 CZK/transaction” is actionable. “Azure cost 500K” isn’t.
Continuous FinOps¶
Monthly cost review: 1. Anomaly detection — What changed compared to last month? Why? 2. Waste report — Unused resources, oversized instances 3. Optimization recommendations — Specific actions with estimated savings 4. Forecast — Projection for next 3 months 5. Action items — Who does what by next review
Časté otázky
30-50% compared to unoptimized state. Quick wins (right-sizing, unused resources) in the first month. Reserved instances and architectural changes in 3-6 months.
Not to start with. FinOps principles (visibility, optimization, governance) can be implemented by the platform team initially. A dedicated FinOps role makes sense from ~500K CZK/month cloud spend.
One-time optimization saves for a month. FinOps is a continuous process — monthly reviews, automatic alerting, team ownership. Savings are maintained and grow.