Skip to main content

Enable anomaly detection

Enable anomaly detection on a Snowflake table to automatically monitor row count and freshness. Once enabled, Snowflake's ML model learns your data's normal patterns and flags deviations—no manual thresholds required.

Prerequisites​

  • Data Quality Studio must be set up for Snowflake and enabled on the connection.
  • The MANAGE_DMF stored procedure must include anomaly detection support (the ANOMALY_DETECTION parameter). If you set up Snowflake before anomaly detection was available, upgrade your stored procedure first.
  • You need permission to create rules on the table.
  • The table must be a Snowflake-connected asset.

Enable anomaly detection​

  1. Navigate to the table where you want to enable anomaly detection.

  2. Open the Data Quality tab.

  3. In the summary section, locate the Anomaly Detection toggle.

  4. Turn the toggle on. A confirmation dialog appears explaining that anomaly detection requires approximately two weeks of training before results become available.

  5. Click Confirm to enable.

Atlan creates two rules automatically:

  • Anomaly Detection - Row Count: monitors the number of rows in the table
  • Anomaly Detection - Freshness: monitors how recently data was updated

Both rules appear in your rules list alongside any existing rules.

Monitor training status​

After enabling, the rules enter a Training state while Snowflake's ML model collects data and learns your table's patterns.

What you seeWhat it means
Training (~2 weeks)The model is learning your data patterns. No results yet.
ActiveTraining is complete. Results are flowing and anomalies are being detected.
ErrorSnowflake failed to enable anomaly detection. Toggle off and on again to retry.

You don't need to take any action during training. The status transitions to Active automatically once the first anomaly detection results arrive from Snowflake.

View anomaly detection results​

Once active, anomaly detection results appear in the same places as other DQ rule results:

  • Rules list: the two anomaly detection rules show pass or fail status with the latest metric value.
  • Rule detail popover: shows the measured value alongside the ML-predicted forecast and expected range (upper bound and lower bound). If the actual value falls outside this range, the rule is marked as failed.
  • Alerts: if you have alerts configured, anomaly detection failures trigger notifications through your configured channels.

For details on each result field—including forecast, upper bound, lower bound, and the is-anomaly flag—see Anomaly detection results.

Disable anomaly detection​

  1. Navigate to the table's Data Quality tab.

  2. Turn the Anomaly Detection toggle off.

This removes both anomaly detection rules and disables ML-based monitoring on the table's DMFs in Snowflake. Historical results are preserved.

Need help​

If you have questions or need assistance, reach out to Atlan Support by submitting a support request.

See also​