Notifications
Monte Carlo's framework for Notifications makes it easy to send the right alerts to a broad variety of different recipients, including Slack, Microsoft Teams, Email, PagerDuty, Webex, OpsGenie, Jira, ServiceNow, and webhooks.
Audiences
To send notifications, first create an audience. An audience is a named group of one or more recipients. For example, an audience could include just a single Slack channel. Or it could include two Slack channels, five email addresses, a Microsoft Teams channel, and a PagerDuty routing key.
Audiences can either be created from within Settings > Notifications, or from the Send Notifications section of monitor creation. They can be easily edited to add, remove, or change recipients.

The modal to create a new audience from within creating a custom monitor.
Adding notifications to an audience
Once an audience is created, you can add it to any existing monitors, by setting their Audiences or Failure audiences properties.
To receive notifications about job failures (Airflow, dbt, Databricks, etc.), add the audience to a job monitor.
Routing alerts by triage priority
By default an audience on a monitor is notified about every alert that monitor raises, as soon as it fires. You can instead narrow an audience so it is only notified about alerts at certain triage priorities.
In the Send alerts to section of monitor creation, each audience has four toggles: High, Medium, Low and Not triaged. A row with all four on is unconditional — that audience is notified immediately, exactly as before. Turn any toggle off and the audience becomes conditional — it waits for Bayes triage to score the alert, and is notified only if the score is one it covers.
A typical setup routes urgent alerts to an on-call channel and everything else to a team channel:
| Audience | High | Medium | Low | Not triaged |
|---|---|---|---|---|
oncall | ✓ | |||
data-team | ✓ | ✓ | ✓ |
Things worth knowing before you narrow an audience:
- Unconditional audiences are notified immediately. Conditional ones are held until the alert has a priority, so they arrive later — usually within a minute or two.
- Not triaged is its own priority, not an error. An alert lands there when it never gets a score: automatic triage is off, triage could not run, or the wait expired. Some audience must cover Not triaged, or those alerts would reach nobody. Monte Carlo will not let you save a monitor that leaves it uncovered.
- Leaving a scored priority uncovered is allowed. If you route only High, you are choosing that Medium and Low alerts notify nobody on that monitor. You will see a warning, not an error.
- Failure audiences are never conditional. A monitor that fails to run produces no alert to triage, so those notifications always go out immediately.
Narrowing an audience only affects the monitors you narrow it on. The same audience can be unconditional on one monitor and High-only on another.
Priorities can change after the alert is sent
Bayes re-assesses an alert when it changes, which can move its priority after it has already notified someone. Notifications follow the priority upwards only:
- If an alert is re-assessed into a higher priority, audiences for the new, higher priority are notified.
- If it is re-assessed into a lower priority, nobody new is notified. Someone has already been paged, so Monte Carlo does not page a second, wider group afterwards.
- Either way the original notification is updated in place, so the priority shown on it always reflects the latest score.
Receive monitor lifecycle notifications
You can be notified when monitors are enabled, disabled, or deleted by enabling theReceive notifications when a monitor is enabled, disabled, or deleted setting on any Slack, MS Teams, or email recipient channel when creating or editing an audience. From then on, any monitor with that audience applied will send monitor lifecycle events to that channel.

Enable monitor change notifications in the Audience creation/edit form
Receive recovery notifications
With Receive notifications when anomalies are normalized turned on for a recipient, Monte Carlo sends a follow-up when an anomaly alert recovers. It's on by default for Slack, Microsoft Teams, email and FireHydrant recipients. Webhook recipients always get recovery notifications. Other recipients, such as PagerDuty, Opsgenie, Jira, ServiceNow and Google Chat, don't get them. A recovery notification is only sent if the whole alert recovers within 7 days of being created.
Customizing notification content
Monte Carlo sets the layout, subject line and branding of alert notifications, including emails, and they can't be changed. To add your own text, such as context or next steps, use the monitor's Notes field. Its text is included in that monitor's alert notifications.
Viewing all your audiences
You can view all your audiences and their monitors in Settings > Notifications.
Click on any of the numbers in the Monitors column to see details of the associated monitors.
GraphQL API
To create and manage new notifications, you should use the Audiences API.
Here's a simplified guide:
Create or Update Audiences:
- Use createOrUpdateAudience to define an audience by specifying its label and notification settings.
- Delete an audience with deleteMonitorLabel.
- Manage individual recipients within an audience using createOrUpdateAudienceNotificationSetting and deleteAudienceNotificationSetting.
Configure Audience Routing Rules:
- Use createOrUpdateAudienceRoutingRule to link a notification rule to an audience. This operation's parameters align with those in the rules field of createOrUpdateNotificationSetting.
- Remove a rule with deleteAudienceRoutingRule.
Details and edge cases
When a recovery notification isn't sent. A recovery notification needs every detection on the alert to normalize. Freshness detections, for example, normalize when the table updates again. A volume change detection records a single change in row count and never normalizes, so an alert that includes one doesn't get a recovery notification.
Updated 6 days ago
