Predicting Program viewership


The client is a Data aggregator for MSO’s in North America.


Problem Statement


Channel or Program rating is not sufficient for optimal ad pricing in targeted advertising. The client wanted to create quarter hour projections of viewership at the household level per network.




Viewership data of 6 months for over 7 million households were used as historical data.


Used IBM Netezza and AWS Redshift as a high-performance database.


Used proprietary analytics framework built on Python for prediction using Machine learning




Predicted viewership on average of 86% accuracy. Used predicted values as input for Media planning platform to determine ad pricing. This is expected to increase revenues for the MSO by around 15% in the first year