Almost every “our Snowflake bill tripled” story has the same shape: nothing dramatic happened. No single runaway query, no one big mistake. Instead, five small inefficiencies compounded quietly until finance noticed. The good news is that the same five levers fix most of it — usually without touching a line of production logic.
1. Idle warehouses billing for nothing
Snowflake bills compute by the second while a warehouse is running. If your auto-suspend is set to 5 or 10 minutes, every warehouse is being paid for long after the last query finished. Drop auto-suspend to 60 seconds (with auto-resume on, so users never wait) and watch a surprising amount of cost evaporate. This one change is frequently the single biggest win.
2. Warehouses sized “just in case”
Teams scale a warehouse up when one query is slow, then never scale it back. A Large warehouse costs 4× a Small for every second it runs. Right-size down by default and only scale up for queries that genuinely spill to disk. Separate warehouses by workload (ELT vs. BI vs. ad-hoc) so one heavy job can not force everyone onto expensive compute.
3. The same expensive queries, over and over
Pull your top credit-consuming queries from ACCOUNT_USAGE.QUERY_HISTORY, ranked by execution time × warehouse size. It is almost always a short list — fix the top ten and you have addressed the majority of avoidable compute. Lean on the 24-hour result cache, and stop the cache-busting patterns (non-deterministic functions, needless churn) that defeat it.
4. Multi-cluster solving the wrong problem
Multi-cluster warehouses solve queueing, not slow queries. If you turned it on to make things faster, you may just be paying for more clusters to run the same inefficient work. Set min-clusters to 1, use an economy scaling policy, and reserve scale-out for genuine concurrency spikes.
5. No guardrails
Resource monitors with notify-and-suspend thresholds turn a surprise quarterly bill into a Tuesday-morning alert. Add object tags so you can attribute spend by team, and review WAREHOUSE_METERING_HISTORY for fifteen minutes a week. What gets measured gets optimized.
Most “out of control” Snowflake bills are not out of control. They are un-watched. Visibility is the fix that makes every other fix stick.
Want the full version with the exact SQL? Our course 101 Cost Optimization Techniques in Snowflake goes deep on all of this, plus an ACCOUNT_USAGE forensic pack and a full audit capstone. Browse the courses, or book a free cost review and we will walk your account together.