Snowflake almost never fails. Queries finish, dashboards load and the nightly dbt run turns green, which is exactly why nobody notices when the account's usage quietly grows month after month. The usual safety nets, resource monitors, auto-suspend and the odd cost dashboard, are built for failures and quotas, not for slow, quiet changes in behaviour.

digna is a Data Quality & Observability platform, and one of the things you can watch with it is the platform your data runs on. This video takes six beliefs many Snowflake teams hold about their bill and checks each one against the account usage views of a fictional pharma company in digna's demo project.

It's all in account usage

Snowflake already records everything you need in SNOWFLAKE.ACCOUNT_USAGE: warehouse metering, query history, warehouse load, serverless credits, storage, task runs and logins. digna reads those views where they live and learns what normal looks like for every warehouse, every dbt model and every report, including the weekly rhythm and the month-end close.

Six beliefs, and what the data says

"A resource monitor will tell us." It will, once a quota is used up. A transformation warehouse that grows a little every week never trips anything: in the demo it went from about 17 credits a day in February to more than 30 by July without a single red day, and digna's analytics put it at the top of the trend ranking long before any quota would have.

"A bigger warehouse just runs faster." It also bills by size. When the analysts' warehouse was resized from small to extra large for a demo and nobody scaled it back, its credits per active hour went up more than sevenfold overnight, and digna flagged it that same day.

"Auto-suspend takes care of idle time." Until somebody changes it. When the analysts' auto-suspend was raised to sixty minutes, one simple rule on the daily warehouse snapshot, no warehouse above ten minutes, caught it the same day.

"If a query finished, it's fine." A cartesian join ran for six hours and scanned about ten terabytes on a warehouse that normally reads half a terabyte a day, and three weeks later a report lost its partition pruning and scanned almost three times its usual data. Both finished, and both were flagged.

"Serverless looks after itself." It does, and it bills for it. When search optimization was switched on for large sales tables, its credits jumped more than eightfold in a day, and when a new semantic model started generating huge SQL, cloud services on the reporting warehouse went from about 10 to 60 % of its compute.

"If something breaks, we'll hear about it." Sometimes. Twelve of thirty-seven dbt tasks failed on the same compilation error one night, a change to a sales view broke 176 dashboard queries, and one day brought 238 more failed logins than the day before. digna caught all three with simple rules on the account usage views.

For the metrics behind these checks, see how to track Snowflake usage, cost and performance.

What digna leaves alone

Just as important is what doesn't get flagged. At month end the finance warehouse uses almost half as much again as on a normal day, and because digna has learned the closing calendar, it stays green, while a fixed alert would fire every single month.

One page in the morning, or one question to your assistant

All of it lands on one page every morning, green almost everywhere and red only where something moved overnight. And if you'd rather ask than look, ask your AI assistant which Snowflake warehouses used more credits this year: with the digna skill it asks digna's MCP server and answers from digna's own results, warehouse by warehouse, without ever reading the data itself.

Everything runs inside your own environment, right next to the account usage views it reads, so nothing ever leaves your organisation.

Book a demo and see it on your own Snowflake account.

Frequently asked questions

Which Snowflake views does digna use?

In the demo: WAREHOUSE_METERING_HISTORY for credits per warehouse and active hour, QUERY_HISTORY for runtime, bytes scanned and failures, WAREHOUSE_LOAD_HISTORY for queueing, METERING_DAILY_HISTORY for serverless credits, storage, TASK_HISTORY for dbt tasks and LOGIN_HISTORY for failed logins, plus a daily snapshot of SHOW WAREHOUSES.

Isn't a resource monitor enough?

A resource monitor acts when a credit quota is used up. It doesn't tell you that one warehouse has grown 80 % in five months while every day looked normal; digna's trend ranking does, long before the quota is reached.

How does digna notice a forgotten warehouse resize?

It tracks credits per active hour for every warehouse. A size step doubles that number, so a resize shows up the same day as a jump against the warehouse's learned baseline, independent of how busy the warehouse was.

Will month-end and quarter-end peaks trigger alerts?

No. Seasonality is part of the learned baseline, so a finance warehouse that uses half again as much at month end stays green, while the same jump on an ordinary day would be flagged.

Is Snowflake the only platform digna can observe?

No. Observing the data platform is one use case of digna, the Data Quality & Observability platform; the same approach works for Databricks, Teradata and other warehouses, alongside monitoring the data itself.