Pega Customer Decision Hub notifications

The action alerts feature triggers notifications in the main portal of Pega Customer Decision Hub whenever an algorithm detects a significant change in the performance of established actions, such as unusually high or low impressions, and an unusually high or low click/accept rate, or whenever the number of daily impressions of new actions is too low.

Notifications are visible in the top right corner of the Pega Customer Decision Hub portal. The number next to the notification icon indicates the number of new (unseen), system-generated daily alerts. Click the notification icon to expand the list of action alerts, sorted by date, with the most recent alert at the top of the list. For established actions, the notification will be prefaced by the date when the spike or drop occurred, the reason for the alert, the KPI it affects, and the time stamp when the alert was generated. For new actions, the notification is triggered when an action logs fewer than a set number of impressions, alongside the alert generation time stamp, and the date when the algorithm detected low performance.

Click an alert to open the Action Performance landing page, prepopulated with action data from the selected alert in a daily view, with the date of the alert as the ending date. A banner is visible at the top of the action details when you click the notification, reinforcing the alert and highlighting the details of the detected change in performance.

Alert algorithm overview

Alerts are generated using an ongoing anomaly detection algorithm that monitors the production data of established actions for changes in impressions and click and accept rates, as well as new actions for low impression rates. Alerts give you an automated insight into any significant changes in the performance of your actions. Notifications are generated daily, as the data is uploaded into the system. There might be some differences, depending on your system configuration.

For established actions, a global threshold parameter defines when unusual activity triggers an alert (by default it is set to 30%). Therefore, the alert is triggered if:

• Current weekly activity is at least 130% of the previous week.
• Current weekly activity is at most 70% of the previous week.

The algorithm is a rolling 7-day-window activity algorithm, which means that it compares data for the current 7-day period to the previous 7-day period, which helps to smooth day-to-day variances, enables daily monitoring, and ensures that the alerts are timely when an anomaly is established. The initial ramp-up period needed for the algorithm to collect enough data is 14 days. Error propagation accounts for additional uncertainty when monitoring rates, such as click and accept rates.

Actions that are newly deployed to production are monitored as new actions. The algorithm monitors the number of daily impressions against a configurable threshold value that you can configure, so that you can track new actions and ensure that they are given sufficient opportunity to perform. Out of the box, an action is considered new for a period of 28 days. On the 29th day, the algorithm will begin to monitor the action for unusually high or low activity, as described above.

The monitoring algorithm looks for the date on which the action is first deployed, to determine whether the action should be monitored as new. If the start date is specified for the action, the algorithm will use this date. For more information about action attributes, see Availability. If there is no start date, the algorithm will look for the earliest date that the action was moved to production. If this date is not available, the algorithm will use the date when the action was created.

For example, you can set the period for monitoring of new actions to 14 days. Then, actions are considered new for the period of 14 days, counted from the start date or the date on which the action was first deployed.