To configure a report to detect anomalies in data, select the metrics that you want anomaly detection to run on. You can select the metrics from the dashboard or from the report itself.

To select metrics for anomaly detection, the anomaly detection feature on the Company Settings page must enabled and a holiday list file selected.

Complete the following procedure to select metrics for anomaly detection.

1. Go to the dashboard that has the report that you want to enable anomaly detection on.
2. Open the report.
3. Click the hamburger menu and select Actions > Set Anomaly Detection Metrics.
   1. Select the metrics on which to apply anomaly detection and set how frequently (hourly versus daily) that anomaly detection runs for each metric.

      Hourly detection cannot exceed 25% percent of the allowed metric quota for anomaly detection. For example, if your organization is at the maximum allowed number of metrics for which anomaly detection can run (20 metrics), hourly detection is allowed for only five of the metrics.

      If you set anomaly detection to both hourly and daily for a single metric, that counts as two separate metrics.

   2. Optional: Enable the Alert feature, select the alert method, and enter email of additional alert recipients.

      To use Slack as an alert method, you must configure Slack integration on the Company Settings page.

      You can use the Alert Manager to manage and modify anomaly detection alerts.

   3. Click Apply.

   An indicator is displayed at the top of the report, showing the progress of anomaly detection as it is applied to the metric. When anomaly detection is fully applied to the metric, the progress indicator goes away, and a View by option is added to the report interface.

   Note: It can take some time for anomaly detection to collect historical data and to calculate data deviations for the specified metric. In most cases, for the initial anomaly detection calculations, data is made available in ten to thirty minutes. After the initial calculations, the system just needs to calculate for the new dates, and that takes less time. In a worst-case scenario, for example when a single metric with many anomaly points and with metric dimensions that have many distinct values, anomaly detection can longer than 30 minutes display data.

4. Select View by > Anomaly to open the report in the Anomaly Detection view.

   You can select the report metrics from the METRIC list to view anomaly detection data for a specific metric.

   For metrics configured for both daily and hourly anomaly detection, you can toggle between views.

   Metric data displayed in the green shaded area in the chart is considered normal based on historical precedent.

   Anomalous spikes or dips in data are displayed as red dots above or below the green shaded area. These red dots represent the anomaly data points for the metric.

5. To view information about the factors that contribute to the anomaly, place your cursor over the anomaly data point and click View contributing factors.

   A table for the chart is populated with information about contributing factors.

   Dimensions and Dimension values are grouped by the contributing metric.

   Click Suggestion in the Action column of the table to explore sessions where deviations for the metric occurred.

   The Suggestion link shows the Segment details (the dimensions and dimension values) for sessions where there was a higher or lower than expected value for the metric being reported on.

   Use the information from the Suggestion link to explore sessions where anomalies occurred in a more detail.

   When you click Search aggregated sessions, anomaly detection searches for sessions with the dimensions and dimension values listed in the Segment details. If sessions are found, you are taken to Session Search, where a list of sessions for the related segment is presented. From the list, you can replay sessions to try and determine the root cause of the anomaly.

   Because some out of the box events (like hit count and session count) occur on every session, anomaly detection searches sessions where the metric DOES NOT occur on the specified segment. In such cases, anomaly detection suggests checking other factors that might be contributing to the down deviation.

   There might be cases where a customized event, like successful checkout (count), has a down deviation for the target metric 'revenue'. If the analysis of contributing factors reveals that a successful checkout count from visitors that use Firefox, has a down deviation for revenue, anomaly detection will suggest searching sessions that do not have 'successful checkout event' on Firefox. The logic being that those sessions might provide some insight on the issues preventing visitors from completing the checkout.