Big Data Moscow 2018


Moran Gavish

Outbrain Inc, Israel


Moran Gavish is a Senior Data Scientist at Outbrain, the world’s leading content discovery platform.
His work focuses on research and development of predictive models and metrics for Outbrain’s recommendation system.
Prior to joining Outbrain, Moran served as a Researcher at the Machine Learning Technologies Group of IBM Research Labs.
And prior to that, Moran led the data science research in a company aimed at predicting stock markets behavior.
Moran holds MSC. in Computer Science and a Master in Business Administration from the Technion – Israel Institute of Technology.


Look-a-likes – How Internet Giants Reach the Most Relevant Users at Scale

In recent years, user-targeting has become an important tool for online marketers and web marketing platforms aiming at marketing optimization. A marketer provides the marketing platform with a list of specific targeted users (“seed audience”), and the marketing platform (Google, Facebook, Criteo, Outbrain, etc.) presents the marketer’s campaigns only to these users. There are various criteria by which marketers select users to be part of their seed audience. For example – users who have previously made a purchase in the marketer’s online store; users who have previously made a purchase in its brick-and-mortar store; users who started, but did not complete a purchase (“retargeted users”), users who have visited the marketer’s website, registered to a newsletter; or users with specific interests or demographics the marketer identifies as important. Targeting users based on past behavior has proven to be extremely predictive of the likelihood of these users to convert at low marketing costs. However, for most marketers and businesses, the seed audience tends to be relatively small. Thus, while the ROI on user-targeting is very high, it may be quite limited in volume. In this talk, I will present a lookalike modeling technique, during which the marketing platform takes the marketer’s seed audience, characterizes their online behavior using the data available to the platform’s operator, and then presents the marketer’s campaign to “similar” users, called “lookalikes”. It therefore allows a marketer to amplify their reach to relevant users by several orders of magnitudes.

Date: October 11, 2018