How Anomaly Detection works

The goal of Anomaly Detection is to find transaction profiles with the largest decrease in transaction success rate over a period of time. These profiles are called anomalies.

A transaction profile is the data associated to the transaction: gateway name, payment type, operation type, card country, card scheme, etc. There are many variables, each able to take on many values. We have therefore so many combinations of transaction profiles that we need an algorithm to explore the profiles in an intelligent way.

ProcessOut evaluates the performance of every transaction profile with sufficient volume daily. We benchmark performance by comparing each profile with its performance in the past. For example, if Anomaly Detection runs on February 15 to detect anomalies for the past week (February 8–14), we will compare performance against the week before, i.e. February 1–7. The transaction success rate rate variation is the difference in conversion rate between the analyzed period and the benchmark period.

To summarize, the algorithm explores the transaction profiles guided by the previous metric (variation of transaction success rate) to find the profiles with the highest decreases, which are anomalies.

Anomalies

For each anomaly, Anomaly Detection provides the transaction profile associated, the impact in terms of volume and amount, the main issuer for this profile and the main decline reasons. The amount impacted is calculated based on the average transaction amount for the anomaly’s transaction profile.

In order to detect transaction success rate drops at different scales, we provide analysis of three period types:

• One week (default)
• One month
• Three months

Anomaly Detection on the Dashboard

You can switch the analysis period by clicking on the Analysis period button in the top right corner.

The algorithm runs every day so we recommend to check the Anomaly Detection page regularly, in case there are new anomalies detected.

Overview page

The top box of the Overview page presents Anomaly Detection with two statistics for the selected analysis period :

• The number of anomalies detected
• The amount impacted by these anomalies

Anomaly Detection page

This feature has its own page and gives insights for each anomaly. There is the date of the last update to know if the anomalies are recent or not.

Each row of the table represents an anomaly. By clicking on a row, it displays the transaction profile and two graphs:

• A view of the transaction success rate decrease
• The distribution of the main decline reasons