Using data science to find hit TV programmes

Data science is transforming the way we watch television  

Providing previously unparalleled customer insights, increasing security and predicting future trends: these are just a few of the ways data science and machine learning are reshaping business.

The television industry is one of many areas of business undergoing a dramatic shift, with TV viewing data offering an abundance of potential insights into how people interact with streaming platforms andthis data is already changing the way television is commissioned and watched.

An Imperial Data Spark project, delivered as part of the Imperial Data Science Intensive programme and carried out in partnership with media tech company Digital-i, a technology and insight company specialising in TV and media industry analysis, researched several key questions:

1. What classifies as a subscription video on demand (SVOD) hit?

2. Which qualities mean a series or movie is likely to perform well on SVOD?

3. What is the most effective way of choosing which titles could perform well on a streaming platform from a catalogue of broadcaster content?

Understanding what makes a hit

Now more than ever, it is important for streamers and broadcasters to know what defines a hit. However, with streaming platforms notoriously secretive with their data, services only have access to their own performance ratings. This makes it difficult to know how content and platforms are performing comparatively.

There are also challenges when it comes to comparing streaming ratings with traditional television figures. For example, while 10 million may be a good number for a prime-time terrestrial viewing slot, how does this compare against a series that got 10 million views of the first episode within a month on Netflix?

Streaming platforms need to define parameters for a streaming success and overall performance not simply based on their own data, but by understanding how audiences interact with competing platforms.

The students’ innovative approach

Using data from Digital-i’s analytics tool (SoDA) and linear TV schedule data, the team were tasked with creating a proof of concept which was able to produce a “hit rate” metric (% chance of a show being a hit). The students proved that, with the right data inputs, a bit of tweaking and wider datasets, it is possible to predict, with some degree of accuracy, which titles will perform well on SVOD services. 

The model was able to accurately highlight shows that were hits on streaming platforms other than Netflix with a high success rate. The fact that they were able to tackle industry problems from a fresh angle and achieve such promising results in a short time, proves the validity of the project.

In our minds this project creates a way for local broadcasters to maximise and optimise the success of their quality content by mining their back catalogues for potential SVOD gold,

says Ali Vahdati, Founder and CEO of Digital-i.

It will give streaming platforms the chance to acquire domestic content that will appeal to their subscribers.

This insight could also give broadcasters which produce and co-produce their own content a revenue boost and have the added effect of increasing awareness of their brand among “hard to reach” audiences that principally consume content on streaming platforms.

Team members: Cintia Millan, Alasdair Charlton-Jones, Alex Inch, Lars Herberholz (Imperial Data Science Intensive 2021)

Academic Mentor: Jose Nunes Teixeira Vaz Moreno