Snowflake FinOps: How Do You Reduce Snowflake Costs Without Slowing the Business Down?
Snowflake FinOps starts with a simple fact: most Snowflake overspend comes from warehouses running with no one using them, compute sized for jobs that no longer exist and costs nobody owns. A Snowflake health check finds them in weeks, without new tools.

You reduce Snowflake costs by finding the compute that runs without serving the business, fixing how warehouses are sized and switched off, and giving every cost an owner who can explain it. In our experience most of the savings come from configuration and ownership, not from new tools, and a focused health check can find them in a few weeks. That combination of technical fixes, clear ownership and shared cost measures is what FinOps for Snowflake means in practice.
In short:
- Snowflake charges for compute by the second while a virtual warehouse runs, so idle and oversized warehouses are the most common source of overspend.
- The usual causes are cautious auto-suspend settings, one warehouse shared by every team, full table rebuilds and dashboards that refresh far more often than anyone reads them.
- Technical fixes only last when every warehouse has an owner and finance and the data team agree on one cost measure.
- In one Cadvel assessment for an international retail group, the first six weeks identified around 40,000 dollars a month in potential cloud savings on a Snowflake platform.
- A Snowflake health check covers five areas: where the credits go, configuration, data movement and storage, ownership and cost reporting.
Cost control has become one of the most common reasons companies contact us. Clients come to Cadvel with a Snowflake bill that grows faster than their business and a data team that can explain each line of it technically but cannot say which part is actually worth the money. Helping them answer that question, and then bring the bill back in line, is one of the core areas we work in, and the patterns are similar enough across industries that I want to describe them here.
Why do Snowflake costs grow faster than the business?
Snowflake costs grow faster than the business because compute is billed for every second a warehouse runs, and most companies never check whether those seconds were useful. A virtual warehouse consumes credits while it is switched on, whether or not it is doing useful work, and each step up in warehouse size roughly doubles the credits it uses per hour. Speed is often worth paying for, because a report that reaches the management team on time has real value. The cost becomes a problem when nobody checks which of those running hours actually served the business and which simply passed because a warehouse was left on or sized for a job that no longer exists.
That is also why the bill is hard to read. On the invoice, a credit spent on the month-end close looks exactly the same as a credit spent on a test environment that nobody remembers creating. Snowflake records in great detail what ran, when, on which warehouse and for how long, but until that information is connected to teams, reports and processes, the bill stays one number that goes up a little every month.
Where does Snowflake money usually go to waste?
Snowflake money usually goes to waste in a handful of places, and almost all of them come down to compute running without a clear business reason. These are the patterns we find most often in client platforms:
1. Idle warehouses. A warehouse keeps running long after the last query finished, because its auto-suspend setting was never changed from a cautious default.
2. Oversized warehouses. A large warehouse handles small, routine reports because it was sized once for a heavy job and never resized.
3. Shared warehouses. Loading, reporting and data science all run on one warehouse, so nobody can tell which team drove the cost and one heavy job slows everyone else down.
4. Full rebuilds. Large tables are rebuilt from scratch every hour when only a few thousand rows changed.
5. Over-eager refreshes. Dashboards refresh every fifteen minutes for reports that someone opens once a week.
6. Forgotten copies. Old environments, clones and long history retention keep adding storage and maintenance cost after they stopped being useful.
Each of these is small on its own. Together they are the difference between a platform that grows with the business and one that grows faster than it.
Why does Snowflake cost optimization need FinOps, not just technical fixes?
Snowflake cost optimization needs FinOps because technical fixes alone don't keep costs down: the same patterns return unless three business questions are settled. A good data engineer can apply most configuration fixes in days, yet six months later the bill often creeps back, because nothing changed in how the platform is owned and measured.
The first question is who owns each cost. A warehouse without an owner never gets questioned, and switching it off feels riskier than paying for it. The second is what each cost is for, which means tying workloads to the teams, products or processes they serve, usually through separate warehouses per workload and consistent tagging. The third is which number everyone agrees on. When finance, the data team and the business units each produce their own report on platform cost, every meeting becomes a debate about whose figure is right, and the decision about what to do never gets made.
This is why we treat Snowflake cost optimization as a data problem rather than a savings exercise. Once usage data is joined to the business, the bill stops being one number and becomes a set of decisions that managers can actually make, such as whether a dashboard is worth what it costs to refresh.
What can you learn from how Snowflake manages its own costs?
You can learn that even the company that builds the platform starts cost control with a business question rather than a technical one. Alex Landis, a senior finance manager at Snowflake, described in an interview with the FinOps Foundation at FinOps X 2025 how Snowflake runs its own efficiency programme, which the company calls the Tiny House Concept. As Landis explains it, the idea comes from living in a tiny house: you have to decide what the largest space is that you really need to live your life, and everything beyond that is a cost without a purpose. Snowflake applies the same question to its own cloud resources, and it treats the programme as a cross-functional effort sponsored by senior leadership rather than an engineering clean-up.
I find Landis's metaphor useful because it changes where the conversation starts. Most cost reviews begin with the current bill and ask what can be cut, which turns every meeting into a negotiation about what each team is willing to give up. The tiny house begins at the other end, with how much platform the business actually needs, at what speed and freshness, and only then looks at what you are paying for beyond that. That is also how we approach a health check.
What does a Snowflake health check include?
A Snowflake health check is a short, fixed-scope review of how a platform consumes credits, how it is configured and who owns its costs, ending with a ranked list of savings. At Cadvel it covers five areas:
1. Where the credits go.We break the bill down by warehouse, workload, user and time of day, and separate costs that serve clear business needs from those nobody can explain.
2. Configuration. We review warehouse sizes, auto-suspend settings, scaling policies and resource monitors against how each warehouse is really used, which is where the quickest savings usually sit.
3. Data movement and storage. We check how often tables are rebuilt versus updated incrementally, how long historical data is kept and which copies and environments are still costing money.
4. Ownership and attribution. We check whether every warehouse and workload has an owner and a business purpose, and whether costs can be attributed to teams or products.
5. Cost reporting. We agree with finance and the data team on a small set of cost measures, such as cost per data product or per report people actually use.
The result is not a long report full of settings. It is a short list of changes ranked by monthly savings against effort, with quick configuration fixes separated from structural ones. In the [international retail group assessment](https://www.cadvel.com/case-study/international-retail-group-sales-reconciliation), that approach identified around 40,000 dollars a month in potential savings in the first six weeks, because costs could finally be tied to the reports and processes that generated them.
How can you start reducing Snowflake costs this month?
You can start reducing Snowflake costs this month with four steps that need no new tools:
1. Look at cost by business area, not by resource. Ask your data team for last quarter's Snowflake cost split by team or product. If they cannot produce it, that is your first and most important finding.
2. Give every warehouse an owner. One name per warehouse, someone who can explain what it is for and would notice if it disappeared.
3. Agree on one cost measure. Choose a unit the business understands, such as cost per report used by management or cost per order processed, and track it monthly alongside revenue.
4. Fix the obvious before the clever. Idle warehouses, oversized warehouses and full rebuilds of large tables are usually where the first savings are.
Frequently asked questions
How much can a Snowflake health check save?
It depends on how the platform has grown, so we do not promise a percentage in advance. As one reference point, in an assessment for an international retail group the first six weeks identified around 40,000 dollars a month in potential cloud savings.
Will reducing Snowflake costs slow down our reports?
Not if it is done by workload. Most savings come from compute that runs when nobody needs it, and separating workloads often makes important reports faster, because they no longer share a warehouse with heavy jobs.
Do we need a FinOps tool to control Snowflake costs?
Not to start. Snowflake's own usage data, sensible warehouse settings, resource monitors and consistent tagging cover most of what a company needs in the first phase. Tools help later, once ownership and cost measures are agreed.
What is the difference between Snowflake cost optimization and FinOps?
Cost optimization is the set of fixes that reduce spend, while FinOps is the ongoing practice that keeps it under control, with owners, shared cost measures and regular reviews between finance, the data team and the business.
Who should own Snowflake costs in a company?
Each warehouse or workload should have a business owner who can explain what it is for, while the data team owns the configuration and finance owns the overall budget and the agreed cost measures.
A Snowflake bill that grows with your business is a sign that the platform is doing its job. A bill that grows faster than your business, with nobody able to say why, is a sign that part of what you pay for no longer serves anyone. So here is my question for you: if your CFO asked tomorrow which part of the business drives the largest share of your Snowflake bill, how long would it take your team to answer?
Sources: FinOps Foundation, "How Snowflake Optimizes for Internal and Customer-Facing Value with FinOps", interview with Alex Landis, senior finance manager at Snowflake, recorded at FinOps X 2025 and published 23 August 2025; Snowflake documentation on virtual warehouses, credit consumption and resource monitors; Cadvel case study, International Retail Group Sales Reconciliation.



