Anomaly Detection

Anomaly Detection & Smart Mobile Alerts detect important changes in data, send alerts notifying of the anomaly, and provide links to the data where the flag was created. They surface important insights and issues automatically, saving time when analyzing data.

Interested in learning more about how Mixpanel uses machine learning to help surface insights and make data more actionable? Here’s an article that covers Mixpanel’s approach to machine learning.

Anomalies

Anomaly Detection automatically analyzes data and triggers alerts when trends are detected. To start receiving anomalies for a specific report, add the report to "My Dashboard" or a dashboard you have created. Alerts are also generated automatically for any saved report that has been viewed at least twice in the last 60 days.

Anomalies can be accessed from the bell icon in the upper right hand corner of the UI. Clicking on an anomaly from the bell will take you to the relevant report where there will be a bubble highlighting the relevant data.

Mixpanel utilizes anomaly detection in both Smart Mobile Alerts on Mixpanel’s mobile app and Web Anomalies in Mixpanel’s webapp. Mixpanel’s approach to anomaly detection is covered in detail here.

2017-11-18_01_06_22.504612-V2_NotificationsAnomalies.png

Anomaly Explanations

Anomaly Explanations (“Explains” for short) is the latest machine learning model running in Mixpanel’s automatic anomaly detection product.  Whenever Mixpanel detects an anomaly in your data, Explains checks for any segments of users that could be responsible for the unexpected change.  When significant driving segments are found, they’ll be presented with an explanation:

When clicking into the alert, greater detail on the anomaly and its driving segments, with quick links to explore the data further will be shown:

Reports with Anomalies

Anomalies can be detected in the following reports:

Segmentation & Insights

  • Daily & weekly time granularity
  • Totals & Uniques

Funnels

  • Anomalies for the conversion rate of individual steps as well as the overall conversion rate

Retention

  • Anomalies on the % of users retaining on a daily time granularity

The feedback you provide (Useful/Not Useful) is used to inform and improve the machine learning models that power Notifications.

Anomaly Requirements and Details

Anomaly detections will only be visible to individual users, not all project members, who do any of the following:

  • save reports on the 'My Dashboard' report
  • have Funnels & Retention reports saved on Dashboards that they have viewed in the last two weeks
  • have saved Funnel & Retention reports that they have viewed in last 90 days

There are several data requirements needed for anomalies to start to be detected.

Segmentation & Insights

  • Daily granularity - 120 days of data
  • Hourly granularity - 120 hours of data

Funnels & Retention

  • 60 days of data + user selected conversion window

Also note that detected anomalies will be prioritized and up to 3 anomalies will be delivered per day.

How do Smart Mobile Alerts and Web Anomalies interact?

Currently, the same machine learning model powers both Smart Mobile Alerts and Web Anomalies but alerts are separate from each other. For example, if you have a report on the Mixpanel mobile app but not on a Web Dashboard you may receive Smart Mobile Alerts for it but not Web Anomalies. Additionally, if you have the same reports on mobile and web dashboards you will receive alerts in both places and dismissing an alert in one place won’t affect the other.

Is this article helpful?
5 out of 10 found this helpful

Comments

0 comments

Article is closed for comments.