Hover a node · click to focus · ←/→ steps
Labels on every GPU job + gateway tags on every token. FOCUS-aligned showback across AWS / Azure / GCP.
When you need this
- Finance cannot reconcile AWS bill with LLM invoices
- Need multi-cloud AI cost in one schema
- Showback meetings argue about column definitions
- Preparing FinOps practice for board reporting
Prerequisites
- Access to CUR / cloud cost export
- LLM gateway usage export
- Tag taxonomy (team, product, env)
Tools
Requires consistent K8s labels before numbers are trustworthy.
Assess for platform teams; less critical for app-only ML teams.
Deploy as single ingress before adding a second LLM vendor.
Steps
- 1
Adopt FOCUS columns as the contract
Map ChargePeriod, ServiceName, Tags, and BillingAccountId. Agree the AI tag keys with finance once.
- 2
Normalize cloud + GPU allocation
Ingest CUR/FOCUS export; join Kubecost GPU allocation on resource IDs / tags. Fill gaps with estimated allocation rules — document them.
- 3
Normalize LLM SaaS spend
Map LiteLLM/provider invoices into FOCUS-like rows with team tags. Treat tokens as usage units alongside cloud meters.
- 4
Automate the weekly pack
Dashboard + CSV for finance. Top drivers, forecast vs budget, anomalies. Same definitions every week.
Adoption pitfalls
- Custom schemas that finance never adopts
- LLM spend left in spreadsheets forever
- Tags missing → FOCUS export with null dimensions
Adoption checklist
- Tag taxonomy documented and enforced
- Cloud + LLM spend in one weekly report
- Finance signs off on mapping rules
- Anomaly alert on AI spend >X% WoW
SEER REAL assessment / sprint
Assessment scores FinOps maturity for AI. Sprint delivers a FOCUS-aligned weekly pack for cloud GPU + LLM gateway spend.