Data Platform FinOps Attribution

Design a FinOps system for a shared data platform. Chargeback theater that cannot be reconciled to invoices is useless. Do not recommend killing a job until you know which consumer SLO it protects. Cloud and engines: [Snowflake, BigQuery, Databricks, Kafka, object storage] Monthly spend split if known: [compute / storage / egress / streaming] Current tagging or object-ownership coverage: [details] Shared warehouses or autoscaling pools: [details] Cost pain: [runaway queries, idle clusters, storage growth, extras] Constraints: [cannot block ad hoc, must keep freshness SLAs] Create: 1. Cost-object model: team, data product, environment, workload class 2. Attribution method for shared compute (query tags, warehouse-per-team, slot reservations) 3. Storage and time-travel cost attribution including failed experiments 4. Unit economics: cost per successful pipeline run and per queried TiB 5. Guardrails: statement timeouts, auto-suspend, result cache, scan limits 6. Budget, forecast, and anomaly alerts that name an owner 7. Optimization backlog ranked by $ saved vs SLO risk 8. Exception process for month-close and incident forensics 9. Dashboard spec: top offenders, idle spend, untagged usage 10. 30/60/90-day rollout that starts with measurement, not blame Separate avoidable waste from load that exists because a consumer SLA is real. Flag any recommendation that moves cost to another team without reducing it.