How to Audit Your Snowflake Bill from a CSV Export (Feb 2026)

If you only do one Snowflake cost exercise this quarter, do this one. It takes 60 seconds, you don't need a Cost Insights tool, and you'll find ~30% of waste before lunch.

Step 1 — Export your Usage CSV

In Snowflake:

1. Admin → Usage
2. Set the date range to Last 30 days
3. Group by: Warehouse (this is the key)
4. Download CSV (top right of the table)

You can also run this SQL and download the result:

SELECT warehouse_name, 
       SUM(credits_used) AS credits, 
       SUM(credits_used) * 3 AS approx_usd  -- adjust to your $/credit
FROM snowflake.account_usage.warehouse_metering_history
WHERE start_time >= DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY warehouse_name
ORDER BY credits DESC;

Step 2 — The patterns we find in every CSV

1. The "always-on X-Large that runs 2-second queries"

The single most common Snowflake waste pattern. You'll see one warehouse using 30%+ of all credits, but the queries on it are sub-second. The fix: drop to Medium or Small. p95 latency goes up ~300ms, credits drop 4–8×.

2. Idle warehouses (the AUTO_SUSPEND problem)

In your CSV, sort warehouses by credit-hours / query-count. Anything > 0.5 credit-hours per query is suspicious. The cause: AUTO_SUSPEND set to 600 seconds (10 minutes) instead of 60. Drop to 60s — same UX, far fewer idle credits.

3. 30-day time travel on dev databases

Not in the CSV directly — run:

SELECT database_name, retention_time
FROM information_schema.databases
ORDER BY retention_time DESC;

Anything >7 days on a dev/staging schema is paying for storage you'll never use. Drop to 1 day on dev; 0 on temp/scratch DBs.

4. Materialized views nobody queries

SELECT name, schema_name, 
       last_referenced_time,
       DATEDIFF('day', last_referenced_time, CURRENT_TIMESTAMP()) AS days_idle
FROM information_schema.materialized_views
ORDER BY days_idle DESC;

Anything with >30 days idle is paying maintenance cost for nothing. Drop it.

5. Snowpipe or auto-ingest mis-sized

If you see a warehouse like INGEST_WH consuming significant credits but your data volume is small, you've probably got PIPE running on too-large compute. Drop to X-Small.

Step 3 — The 60-second math

Pull the top 5 warehouses from your CSV. Multiply credit usage × your $/credit rate. That's where 80% of your bill comes from. Anything in that top 5 you don't have a good reason for? That's the audit finding.

We've never run this on a >$30K/month Snowflake account and found <$8K/month of savings.

Step 4 — The follow-up SQL

Once you've identified the top 3 candidate warehouses, run:

SELECT warehouse_name, 
       AVG(execution_time) / 1000 AS avg_seconds,
       PERCENTILE_CONT(0.95) WITHIN GROUP (ORDER BY execution_time) / 1000 AS p95_seconds,
       AVG(bytes_scanned) / POWER(1024, 3) AS avg_gb_scanned
FROM snowflake.account_usage.query_history
WHERE start_time >= DATEADD(day, -7, CURRENT_TIMESTAMP())
  AND warehouse_name = 'YOUR_TOP_WAREHOUSE'
GROUP BY warehouse_name;

If the average query is sub-5-seconds, you're over-sized.

Step 5 — Run it through the free auditor

CARTIE's free Snowflake bill audit takes your usage CSV and returns your top 5 credit-leak warehouses, the right-size recommendation for each, and the ALTER WAREHOUSE SQL to apply. No ACCOUNTADMIN access, no IAM, no retention. Drop the file, see the result.

---

If you want the deep playbook, see The Snowflake Cost Optimization Playbook: 7 Patterns That Cut Your Bill 40%.