Before you start
4- A Snowflake account where you can view usage and credit consumption.
- A role that is allowed to change warehouse settings, or a colleague who has one.
- A list of the warehouses your teams and tools use, with an owner for each.
- Basic familiarity with running SQL and reading a query profile.
Snowflake bills for compute through credits, so cost control comes down to three levers: how large each warehouse is, how long it stays running when idle, and how much work each query makes it do. Cost guidance for Snowflake consistently combines right-sizing resources with better query and storage practices, and it starts from monitoring.
This guide gives you one pass through those levers in a fixed order. You measure first, because changing settings without a baseline leaves you unable to tell what helped.
No version number is pinned here, because Snowflake is a hosted service and the cost guidance does not depend on one. Menu labels and exact setting names can differ by interface, so confirm them in the official Snowflake documentation for warehouses before you change anything. This guide does not list exact SQL syntax for that reason.
Step 1 of 5: Measure where your credits go
Open your account's usage and credit views and list credit consumption per warehouse for a recent period of at least a few weeks. You need a baseline, otherwise you cannot prove that a change saved money.
Record three things for each warehouse:
- total credits used,
- the hours of the day when it runs,
- who or what uses it (a team, a BI tool, a scheduled job).
Look for anomalies as well as totals. A sudden spike in a day's consumption usually points to one job or one user, and it is cheaper to fix a single cause than to tune everything. Cost guidance recommends monitoring across all Snowflake services, not only compute, so note storage and other service charges in the same sheet.
Sort the list by credits and work from the top. A small number of warehouses normally account for most of the bill, so that is where effort pays back first.
Step 2 of 5: Right-size each warehouse against its workload
Match each warehouse's size to the work it does, starting with the largest consumer from step 1. Oversized warehouses burn credits on work that a smaller one would finish acceptably, and undersized ones can make queries run so long that the saving disappears.
Work through one warehouse at a time:
- Check what runs on it. Mixed workloads, such as heavy transformations next to light dashboard lookups, are a sign that one size is wrong for part of the work.
- Separate unlike workloads into different warehouses when the mix forces you to oversize for the heaviest job.
- Change the size by one step, in the direction the evidence suggests.
- Run the same workload and compare run time and credits against your baseline.
Change only one setting per warehouse at a time. If you resize and change Auto Suspend together, you cannot tell which change moved the numbers.
The exact command or interface path for changing a warehouse size is not reproduced here. Use the warehouse settings in your account, and check the official Snowflake documentation on warehouses for the current syntax and the available sizes.
Step 3 of 5: Tighten the Auto Suspend setting
Lower the Auto Suspend time on warehouses that sit idle between bursts of work. A running warehouse consumes credits whether or not a query is using it, so idle minutes are pure waste.
Pick the value from your usage data, not from a guess:
- For warehouses used in short, scattered bursts, such as an ad hoc analyst warehouse, use a short idle time.
- For warehouses that serve steady, frequent requests, a very short idle time can cause constant suspend and resume cycles. Choose a value that covers the normal gap between requests.
- For scheduled jobs, make sure the warehouse is not left running long after the job ends.
Some third-party tools automate this by analysing the workload on a warehouse and adjusting Auto Suspend. You can do the same by hand with the hours-of-day data from step 1, which is enough for most accounts.
Confirm that auto-resume behaviour suits your users before shortening the time, so that queries still start without manual intervention. Check the warehouse documentation for the exact parameter names in your interface.
Step 4 of 5: Tune the queries that cost the most
Find the longest-running and most frequent queries on your biggest warehouses, then fix those first. A query that runs for a long time or runs thousands of times a day keeps a warehouse awake and may be the reason you sized it up.
For each candidate, open its query profile and look for where the time goes. Common causes you can act on:
- reading far more data than the result needs, for example selecting every column when a few are used,
- joins that multiply rows before a later filter removes them,
- the same expensive result recomputed many times when it could be computed once and reused,
- scheduled tasks that run more often than anyone consumes their output.
After each fix, rerun the query and compare its run time with the earlier figure. Good query tuning often lets you step a warehouse down a size, which then compounds the saving from step 2.
Cost guidance also lists storage practices alongside query practices. Review tables you no longer use, because storage is billed as well, and handle that separately from compute.
Step 5 of 5: Re-measure and keep watching
Compare credit use per warehouse with the baseline from step 1, after at least as long a period as the baseline covered. Short comparisons are distorted by one-off jobs and weekday patterns.
Write down for each warehouse: the old and new size, the old and new Auto Suspend value, the queries you tuned, and the credit change. If a change made users wait too long, undo it and try a smaller adjustment.
Then set up a routine. Review the per-warehouse view on a fixed schedule, and watch for anomalies so a runaway job is caught in days, not at the end of the billing period. Cost reduction is not a one-off project, because workloads change and a setting that was right last quarter can be wrong now.
Expected result
A sheet with one row per warehouse showing the baseline credit use, the changes you made (size, Auto Suspend value, queries tuned), and the credit use after the change. Credits for the largest consumers are lower, the run time of key workloads is acceptable to their users, and you have a scheduled review in place.
Frequently asked questions
Which should I change first: warehouse size, Auto Suspend or queries?
Measure first, then work on the biggest consumers. Right-size and tighten Auto Suspend on those warehouses, one change at a time, and tune the costliest queries, which can let you reduce size further.
Does shortening Auto Suspend always save money?
Not always. It saves credits when a warehouse sits idle between bursts. If requests arrive frequently, a very short value can cause repeated suspend and resume, so compare against your baseline.
Can I improve costs without touching any queries?
Yes, for a first pass. Warehouse size and Auto Suspend are settings, not query changes. Query tuning is a further lever that often allows a smaller warehouse.
How do I find cost spikes?
Look at credit use per warehouse over time and watch for days that stand out. Cost guidance recommends monitoring across all Snowflake services and identifying anomalies automatically or through regular review.
Where do I find the exact commands for changing a warehouse?
Use the official Snowflake documentation on warehouses. This guide does not reproduce syntax, so check that page for the names that apply to your account.
From Grokem: https://www.grokem.com/data/data-platforms/data-warehouses/how-to-cut-snowflake-costs-right-size-warehouses-set-auto-suspend-tune-queries








