What is linked and quoted below is a fascinating, although very technical, document discussing how Netflix currently studies and makes future content creation decisions based on existing and past viewer data.
While we’ve suggested before (see: Does Netflix Release Too Many Originals? Maybe Ask New CMO Bozoma Saint John) that Netflix seems to throw a lot of their originals at the wall to see what sticks, it’s much more calculated than that. Depending on how geeky and technical you are, the article linked below could be illuminating into their process.
Making great content is hard. It involves many different factors and requires considerable investment, all for an outcome that is very difficult to predict. The success of our titles is ultimately determined by our members, and we must do our best to serve their needs given the tools and data we have. We identified two ways to support content decision makers: surfacing similar titles and predicting audience size, drawing from various areas such as transfer learning, embedding representations, natural language processing, and supervised learning. Surfacing these types of insights in a scalable manner is becoming ever more crucial as both our subscriber base and catalog grow and become increasingly diverse.Supporting content decision makers with machine learning | by Netflix Technology Blog | Dec, 2020 | Netflix TechBlog
To simplify at least some of what I understood — and I’m a very geeky, technical guy — they group story ideas together to find likeness and related types of content. It’s like a Google related search for keywords, using the framework of the stories and then instead of search results, it will yield the type of stories that specific viewing audience will be more likely to watch and enjoy.
Of course it’s not only watching once or few a few seconds, what type of content are viewers repeatedly watching. What are the demographics behind these viewers?
Combine all of these characteristics and an algorithm is developed to be able to predict what types of stories in what age groups in what geographic regions will yield successful results.
None of this guarantees a hit movie or TV show, of course, but it does increase the odds that a high percentage of viewers will watch a certain type of story.
This all reminds me of some fiction writing software that broke down conflicts between characters into types. You could mix and match story types and characters and essentially have a bare bones framework for a type of story. Then you just need to sit down and write the story based on the sequence of actions and events in the computer generated outline. It was all too wooden for me, too structured, that I never used it, but found it interesting.
Does it work? Just saying if you like Diehard, then you’ll like other movies like Diehard? Sure, on a very basic level, it does, but there have been a bunch of Diehard clones and only one Diehard to date. Not even Diehard sequels have been able to catch the lightning in the bottle of the first film.
Studios try to do something similar by remaking and rebooting movies and TV shows that did well in the past. They want to capitalize on the fond fan memories. It does work — until the new production has to stand on its own. Take Cobra Kai, for example, the idea would have been very gimmicky beyond the first episode or two, had it not had its own stories to tell. It did, and fans responded.
In 2020, we’re being studied by computers everywhere. What we search for, what we click on, what we open in email and what we watch on streaming channels. Maybe what will rise from the ashes of all this personal intrusion will be a service that does not study everything we do attempting to meaningfully make decisions based upon it. I mean, is this the future world we want all the time everywhere? Something tells me probably not.