Update to Filtering email bot activity proximity pattern algorithm

[Sreekanth_Reddy]

Hello Nation!

As most of you know, we have launched "Proximity Pattern" a new identification method for email bot activities. Last month, we launched an option to choose either filter or log activities identified as bots. In case you missed it - here is the blog

Proximity Pattern, today identified activities as bot activities based on behavior i.e if multiple activities from an email asset are sent to a specific lead happening at almost the same time i.e <1sec. When comparing two activities following values are considered:

1. Ignore exact duplicates i.e same link, same lead ID, same UA, same IP and almost the same time
2. Attributes used to compare two activities are:
   • lead ID (should be the same)
   • User-Agent (should be the same)
   • IP (should be the same)
   • Email asset (should be the same)
   • Link click or open
   • time difference (should be less 1second)

After a lot of analysis, to improve the sophistication of proximity patterns, we plan to update the algorithm of proximity pattern to consider only the following attributes during comparison:

• lead ID (should be the same)
• Email asset (should be the same)
• Link click or open
• time difference (should be less 1second)

Basically, User-Agent and IP address will no longer be considered in the proximity pattern which means if we are getting clicks almost at the same time from the same lead, same email asset for different clicks under 1second. We have run this analysis on several real data across customers and seen an improved bot activity identification. 

IMPORTANT NOTE: We are going to update the proximity pattern algorithm with this change and launch it in the coming few weeks (part of the Oct 2022 release). Post this update launch, if you have proximity pattern already enabled on your subscription expect that number of activities identified as bot will increase. In case you are not sure of this update's impact on your subscription we recommend you to enable the proximity pattern identification method in log mode, observe and analyze for a few weeks and then choose filter out mode.