Why your Snowflake or BigQuery bill keeps growing, and 9 checks to cut it

Your data bill grows faster than your business. Nobody can quite explain why, and every quarter it creeps up again. This is one of the most common problems we see. It's usually caused by a handful of fixable habits, not by having "too much data". Here are nine checks, written so you can ask your team without needing to be technical yourself.

How these platforms charge you

It helps to know what you're actually paying for:

• Snowflake charges credits for every second a "warehouse" (its computing engine) is running, with a one-minute minimum each time it starts. Each warehouse size up costs twice as much per hour. Storage is billed separately.
• BigQuery, on its default on-demand pricing, charges for the amount of data each query reads. One careless query on a large table can cost more than a day of normal use. Alternatively you can buy reserved capacity.
• Databricks and Redshift mostly charge for clusters while they run, so anything left running overnight costs money.

In every case, the bill is driven by how the data is processed, not just how much of it you have. That's good news, because processing habits can be changed.

Nine checks to cut the bill

1. Are we reloading everything every time?

Many pipelines rebuild whole tables on every run, even when only yesterday's rows changed. Switching to incremental loads, which process only new or changed data, is often the single biggest saving. In dbt, these are called incremental models.

Ask your team: "Do our data jobs reload everything each time, or only what's changed?"

2. Are we paying for idle computing power?

In Snowflake, check each warehouse's auto-suspend setting. A warehouse that waits 10 minutes before switching off bills for those 10 minutes after every burst of activity. Setting it to about a minute for scheduled jobs is common. In Databricks and Redshift, check for clusters that run around the clock.

Ask your team: "Do we pay for computing power when nobody's using it, like nights and weekends?"

3. Are our warehouses or clusters bigger than they need to be?

Teams often size up to fix one slow report and never size back down. Because each Snowflake size doubles the cost, one step too big doubles that warehouse's bill.

Ask your team: "When did we last check whether each warehouse or cluster could be one size smaller?"

4. Do our biggest reports read more data than they need?

In BigQuery, a query that selects every column, or that doesn't filter on the date, reads the whole table. Partitioning tables by date and clustering them on common filters lets queries skip most of the data. Setting a maximum bytes billed limit stops runaway queries. Snowflake has the equivalent in clustering and well-designed tables.

Ask your team: "Are our largest tables set up so reports only read the data they need?"

5. Are dashboards refreshing more often than anyone looks?

A dashboard that refreshes every 15 minutes runs 96 times a day. If people check it once each morning, hourly or daily refreshes will do.

Ask your team: "How often do our dashboards refresh, and how often do people actually check them?"

6. Are we still updating reports nobody uses?

Every report and table that refreshes costs money whether anyone opens it or not. Most platforms log usage, so it's easy to find what hasn't been touched in 90 days and switch it off.

Ask your team: "Which reports and tables haven't been used in 90 days, and can we stop updating them?"

7. Are we copying data we don't use?

Tools like Fivetran charge by how much data they sync (Fivetran bills on monthly active rows). Syncing every table from every app "just in case" adds up. So does loading the same source twice for different teams.

Ask your team: "What are our data tools copying that nobody uses, and are we loading anything twice?"

8. Are test environments full-size and always on?

Full copies of production data for testing double your storage and often your compute. Small samples are usually enough to test changes safely. Storage features like Snowflake's Time Travel also keep old versions of data, which you pay to store.

Ask your team: "Do our test environments need to be full size and running all the time?"

9. Are we on the right pricing plan, and does anyone own the bill?

Once usage is steady, committed or reserved pricing is usually cheaper than paying as you go: Snowflake capacity contracts, BigQuery reservations, Databricks commitments. Few companies revisit this after signing up. Just as important: someone should own the bill and get an alert when it jumps. Snowflake's resource monitors and Google Cloud budgets do this.

Ask your team: "Who owns our data bill, and would we know within a day if it suddenly doubled?"

Want a rough number first? Our free data platform cost calculator turns these questions into a savings estimate in about 60 seconds.

When to get help

If your team can work through this list, do it. Many of these fixes take days, not months. Bring in outside help if the bill is large, the team is stretched, or nobody can explain where the money goes. Our Data Health Check measures your actual usage and gives you a costed list of fixes in two weeks. If you want someone to make the changes, Data Rescue ties part of our fee to the savings we verify.

Common questions