Big Data To Predict Video-On-Demand User Behavior, Driving EUR 4MM Of Revenue

Our client is a major global Telecom player, headquartered in France. They have diversified businesses spanning Mobile, Landline & Internet, Broadcasting, and Music.

Their objective was to improve Average Revenue Per User (ARPU) through targeted customer retention and upsell promotion campaigns. The high-growth VOD and gaming content divisions were identified as focus areas.


  • 60 GB+ dataset with over 450 million transactions
  • Limited existing metadata on movies/series and other VOD content
  • Constantly changing user base for pay-per-view format (low stickiness), with limited subscriber related demographic info (as these are household IDs).

Solution Methodology
  • Observed month-on-month user purchase trends, on various dimensions like purchase frequency, revenue bucket, content price, time and day of the week, etc.
  • Segmented users based on frequency buckets (purchase frequency per month) and their transition behavior (upward, downward, inactive, stable) month-on-month.
  • Augmented content metadata by extracting content attributes (genre, cast, ratings etc.) through APIs from sources such as IMDb and Google Translate for English-French translation
  • Developed machine learning (in this case a Parallel Random Forest model) to predict churn and upsell propensity. Used Variance Inflation Factor (VIF), Mahalanobis Distance and Correlation Matrix to guide feature selection
  • Fine-tuned model performance considering trade-off between accuracy and capture rate – depending on the size of user base and costs associated with intervention efforts.
Business Impact
  • The estimated incremental revenue contribution is to the tune of EUR 4 million per annum, through ahead-of-time prediction of potential churners and potential upsell consumers.

Copyright © 2023 Tiger Analytics | All Rights Reserved