One of Roald Dahl’s many fantastic short stories is “The Great Automatic Grammatizator”, written in 1953 it tells the story of a machine that could write fiction. As I recall it, the machine could churn out books at a phenomenal rate and would incorporate feedback on what proved popular into its future works. We are not quite in the age of Artificial Intelligence (AI) fiction yet, but human data is playing an important part in the way that our entertainment is produced.
TV ratings have long been useful to program makers in optimising their output, but other types of data are now providing more detailed feedback for them to use. The producers of one US TV show decided to use anonymized and aggregated Facebook topic data to understand how their audience engaged with episodes, storylines and characters. In particular, the producers wanted to make the right casting decisions for the next season, as well improve the promotion of the show through advertising and social media.
To do this, the production company set up two different indexes of Facebook topic data. One index looked at their own show and they used VEDO – the intelligence inside DataSift that allows you to organize human data based on its meaning – to isolate references to key episodes, storylines, quotes, memes and characters. The other index measured engagement with breakout actors from recent film festivals and top TV shows. Overall, the production company was able to use data from 1.5 million interactions to inform its decisions.
The first thing the production company was able to determine was which characters were resonating most with people on Facebook, giving great insight into which characters are most popular in the show. It also informed the producers as to which should potentially be retained for future seasons. The producers were also able to see how the engagement around characters changed across different demographic groups (age, gender and state) and they could use this data to help promote the show to these groups. For example, the protagonist’s rival love interests polarised opinion in Wisconsin and Colorado, with one popular in the Midwest and the other in the Rockies. This meant that the most popular character could be used to promote the show in each state.
It was also important to the producers to be able to understand which episodes and scenes were driving the most engagement. In this age of streaming and catch-up TV, they didn’t want to do this just by the time of broadcast. That’s where using advanced classifiers to identify the stand-out scenes came into its own. Armed with this data, they were able to see which of their storylines were generating the most viewer interest in certain demographic groups and use those scenes to advertise the show to that audience. They will also be able to apply that intelligence to future writing decisions.
Finally, the show’s producers were able to take the demographic data they had gathered on the core audience of their own show and see which actors from other shows and films that audience liked. These actors could then be assessed for their suitability to join the show in future seasons.
While it’s reassuring to know that writers and casting directors aren’t going to be replaced by computers just yet, it’s very interesting to see how data helps drive their creative decisions.
To see how other brands made better decisions using Facebook topic data, take a look at our new eBook Busting Brand Myths with Facebook Topic Data.