Anomaly Detection Alerts

Anomaly detection helps you spot unusual changes in time-based metrics and get automatically notified. This is useful when you want to monitor key KPIs without manually checking dashboards.

What Anomaly Detection Does

When enabled for a metric, the system looks at the metric over time and finds values that are unexpectedly higher or lower than usual. It considers:

• overall trend (values going up or down)
• repeating patterns (seasonality), such as weekly or yearly cycles
• sudden changes that do not match the expected pattern

When an anomaly is detected, you receive an alert so you can investigate.

Who Can Use It

You need access to both:

• Alerts
• AI Assistant features

Create an Anomaly Detection Alert

1. Open a dashboard and find a visualization with a trendable metric.

2. Open the visualization menu and go to Alerts, then select Create.

3. In the setup dialog:

   • Select the metric to monitor. If the visualization has multiple metrics, the first metric (in alphabetical order) is selected by default.
   • Select Aggregate by (day, week, month, quarter, year). By default, this matches the granularity used in the visualization.
   • Optionally, add filters (for example Country = US).
   • Select sensitivity: low, medium (default), or high.

4. Choose what happens when an anomaly is detected (notify via email or other available destinations).

5. Confirm to create the alert.

Notes:

• The system uses up to the last 5 years of history by default. If less history is available, it uses what exists.

• If there is not enough data to evaluate anomalies (for example, too few data points), alerts are not triggered until enough history is collected.

• If some data points are missing, the system automatically fills them in using the median value of your available data. This ensures the analysis can continue even with incomplete data.

• Seasonality is calculated statistically and automatically selected. The system uses a single seasonality level that is higher (coarser) than the monitored metric’s granularity. It does not combine multiple seasonalities.

  Mapping used (metric’s granularity → seasonality):

  • day → week
  • week → month
  • month → year
  • quarter → year
  • year → no seasonality

• Date range filters are not used for anomaly detection.

How Often Checks Run

Checks run based on the selected granularity:

• daily checks evaluate the previous day
• weekly checks evaluate the previous week (after the week ends)
• monthly checks evaluate the previous month, and so on

Create Anomaly Detection Alerts With AI Assistant

You can also create and schedule anomaly detection alerts using the AI Assistant.

• The AI Assistant supports the same alert setup workflow as manual alerts (metric, schedule/granularity, destination, recipients, and sensitivity).
• If you do not specify the granularity (or it can’t be inferred), the Assistant asks follow-up questions.
• By default, the AI Assistant uses medium sensitivity. You can change it by explicitly asking for low or high sensitivity.

Example prompts:

• Notify me if there is any suspicious activity on this metric daily.
• Let me know if any anomalies are detected in weekly numbers.
• Set an anomaly alert on Total Sales.
• Monitor Total Sales for anomalies daily.
• Create a scheduled anomaly check for Total Sales.
• Run anomaly detection on Total Sales every day.
• Ping me when Total Sales behaves oddly.

Display Anomalies in Visualizations

You can highlight anomalies directly in a visualization so that unusual data points are visible at a glance on dashboards.

When the Option Is Available

In Analytical Designer, anomaly detection in visualizations is available only when all of the following conditions are met:

• The visualization type is a Line chart
• Exactly one metric is used
• The metric is sliced by a date attribute

Only users with AI Assistant permission can enable anomaly detection in visualizations, both in Analytical Designer and on dashboards.

Configure Anomaly Detection

When anomaly detection is enabled for an eligible line chart, additional configuration options appear:

• Sensitivity

  • Low – only large, rare deviations
  • Medium (default) – balanced signal-to-noise
  • High – more aggressive detection

• Indicator color

• Indicator size

The indicator shape cannot be configured directly. It follows the Distinct point shapes setting in the visualization canvas.

After anomaly detection is enabled, normal data points remain unchanged and anomalies are visually emphasized.

Tooltip and Legend Behavior

When you hover over a highlighted anomaly point, a tooltip explains why the point is marked as anomalous without requiring users to understand the underlying calculation.

When a chart contains a single metric and anomaly detection is enabled, the legend can be shown to help explain anomaly markers. Legend visibility is controlled by the visualization author.

Notes on Alerts and Visualization Consistency

Anomaly alerts can be scheduled even if anomaly markers are not visible in the visualization. In this case, users may receive notifications about anomalies that are not highlighted on the dashboard.

If anomaly detection is hidden for a visualization, a warning can be shown when scheduling an alert, for example:

Anomaly markers are currently hidden on this visualization. Alert recipients will not see anomalies highlighted in the dashboard.

This approach keeps alerting flexible while making the behavior clear to users.

Anomaly Notifications

If an anomaly is found, you receive a notification through the selected channel. The message includes:

• alert name
• metric name
• granularity (how often it is checked)
• time period

• actual value

• applied filters (if any)
• link to open the dashboard or visualization

Example:

• Metric: Net sales
• Frequency checked: weekly
• Period: Week 42 / 2025

• Actual value: 155K

• Filters: Country = US
• Triggered at: 12.11.2025 11:00 CET
• Link: open dashboard