Edinburgh Napier University

  The video games industry is one of the most attractive and lucrative segments in the entertainment and digital media, with big business of more than $150 billion worldwide. A popular approach in this industry is the online freemium model, wherein the game is downloadable free of cost, while advanced and bonus content have optional charges. Monetisation is through micro payments by customers and the focus is on maintaining average revenue per user and lifetime value of players. The overall aim of this research is to develop suitable data-driven methods to gain insight about customer behaviour in online freemium games, with a view to providing recommendations for successful business in this industry.

Three important aspects of user behaviour are modelled in this research - engagement, time until defection, and number of micro transactions made. A multiple logistic regression using penalised likelihood approach is found to be most suitable for modelling and demonstrates good fit and accuracy for assigning observations to engaged and non-engaged categories. Cox’s proportional hazards model is adopted to analyse time to defection, and a negative binomial zero-inflated model results in the best fit to the data on micro payments. Cluster analysis techniques are used to classify the wide variety of customers based on their gameplay styles, and social network models are developed to identify prominent ‘actors’ based on social interactions. Some of the significant predictors of engagement and monetisation are amount of premium in-game currency, success in missions and competency in virtual fights, and quantity of virtual resources used in the game.

This research offers extensive insight into what drives the reputation, virality and commercial viability of freemium games. In particular it helps to fill a gap in understanding the behaviour of online game players by demonstrating the effectiveness of applying a data analytic approach. It gives more insight into the determinants of player behaviour than relying on observational studies or those based on survey research. Additionally, it refines statistical models and demonstrates their implementation in R to new and complex data types representing online customer behaviours.