Mathematics to predict audience ratings

One of the main Italian media companies asked us to realize a support tool for strategic marketing. In order to better plan its schedules and purchases of audiovisual content, the goal of the media company is to predict its prime-time audience on its main networks in the short term.

Description and Benefits

The estimation of the share of a particular content, on a channel, at a specific date and time of airing, is traditionally carried out by experienced professionals. A task based on their know-how and ability to consider a multitude of interconnected variables. This is a process of great complexity, because it must also take into account the performance forecasts of competing channels, therefore requiring many hours of work.
Moxoff has thus developed an estimation system based on mathematical models and machine learning that, in just a few minutes, produces a solid base of share prediction available to the broadcaster’s marketing division, which can then quickly perform various scenario analyses and forecasts on the available strategies.

An ensemble of models for the estimation of the audience index

The methodology applied allowed us to estimate the audience index up to two weeks.

It made use of an ensemble of models, combining different methodologies and algorithms such as Random Forest, XGBoost, Multi-layer Perceptron, and others. To feed the models, in addition to the historical ratings, we used information regarding the characteristics of the content (type, genre, awards won, box office, etc.) structured in a database that is updated every night.

  • Reduction of share estimation time, from 4 hours to 2 minutes
  • Simple framework, to carry out scenario analysis, it is sufficient to change the time slot of one of the contents and the tool elaborates a different configuration from the previous one
  • Accurate Informations, the percentage error is low, equal to 5.8% MAPE (Mean Absolute Percentage Error) on the main network of the broadcaster