Every way we find money left on the table.

Compute is where most Snowflake bills are won or lost. These detectors catch warehouses that idle, run too big, or run when nobody's watching.

• Auto-suspend disabled or too high — warehouses that keep billing while idle; we recommend lowering the timeout (often to ~60s) and can apply it in one click.
• Over-provisioned warehouses — right-sizing down (e.g. from one size to a smaller target) based on real utilization.
• Standard Gen2 candidates — warehouses that would run cheaper/faster on a newer generation.
• Dormant & idle warehouses — never used, or idle for many days, still configured to cost.
• Multi-cluster misconfiguration — multi-cluster on a STANDARD policy, always-on MIN_CLUSTER_COUNT > 1, or a warehouse repeatedly hitting its cluster ceiling under load.
• Off-hours & narrow-window usage — warehouses burning credits outside business hours, and small warehouses running in narrow daily windows that could be consolidated or scheduled.

The same warehouse costs less when the queries on it stop wasting it. These detectors surface the query shapes that quietly burn credits.

• Repeated expensive query shapes — the same costly pattern run thousands of times a month, a prime target for a materialized view, result cache, or rewrite.
• Spilling queries — queries spilling to local/remote storage (measured in GB), a sign of an undersized warehouse or a heavy plan.
• Full-table scans — queries scanning >90% of partitions, where clustering or better pruning would cut the read.
• Queue premium — hours of query queueing that point to concurrency or scaling problems.

Storage is cheaper than compute but compounds silently — retention, clones and staging pile up.

• Time-travel bloat — tables holding large time-travel storage that outlives its usefulness.
• Clone-retained storage — storage kept alive by clones long after they're needed.
• Unqueried databases — databases holding gigabytes that nothing has queried.
• Internal stage storage — data parked in internal stages that keeps accruing.

AI spend is the fastest-growing line on many Snowflake bills — and the least governed. We track it in credits so it doesn't surprise you.

• Oversized models on simple tasks — large Cortex models doing work a smaller model would handle.
• Generative Cortex spend — credits going to generative functions, surfaced and trended.
• Cacheable high-volume functions — high-call-count Cortex functions whose results could be cached.
• Fine-tuning spend — cost spread across fine-tuned models.
• Cortex Search + Analyst — combined credit consumption from Search and Analyst.

The findings that keep spend from creeping back: guardrails, dead accounts, and pure waste.

• Missing resource monitors — warehouses with no spend guardrail at all.
• Dormant users — accounts with no login for a long window.
• Failed-query waste — compute-hours burned by failed or cancelled queries.
• Cloud-services overage — billable cloud-services credits above the free allowance.
• Cross-region / cross-cloud egress — data-transfer cost from replication or reads across regions.
• Snowpipe small-file loads — pipes ingesting many tiny files, an inefficient (and pricier) load pattern.
• Auto-clustering cost — serverless clustering credits weighed against the pruning benefit.