The Client’s primary objective was to develop and execute a bidding strategy for Facebook ads such that ad-serving costs are lowered. They also wanted a targeting strategy based on a set of user demographics which results in significantly high postclick credit card applications.
Facebook allows us to target ads with several dimensions (microsegments) including Age, Gender,Interest, Education, Location, etc. For each of the markets we were running more than 150+ microsegments.
Interpreting the results/performance of each segment and modifying bids and budgets in a real-time fashion was the greatest challenge. The competition in social media is stiff and coupled with that is the low applications to clicks ratio from the Facebook.
We developed a framework for real time bidding for ads in Facebook. The framework identified the best performing audience micro-segments and ads (images), and dynamically came up with optimal bids and budgets for each micro-segment, for each day of the week. The backend algorithms involved a count data based
predictive model. The results of the backend algorithm were communicated to Facebook through an API. The campaign performance was continuously tracked and feedback was used to refine the models in an automated manner. The framework was implemented in Pentaho and the databases used were MySQL and MongoDB.
The number of total applications were in line with the high targets. The cost of applications were below the set target (by 10%) .