Ever started an episode on Netflix with no intention of watching much but spent three hours binge-watching the whole show, just because your next movie needed to be perfect? This isn’t a coincidence; this is data analytics working behind the scenes.

Netflix, YouTube, and Spotify are three of the world’s most famous entertainment platforms, each having its own business model. However, one thing is common in them all. A better understanding of your behaviour gives a better prediction of your future actions.

While Netflix does not simply monitor what movies you’ve watched, the collected data include:

Everything that was listed above goes into algorithms that segment the audience into tiny “taste communities”, not the usual “action fans” category, but rather thousands of extremely narrow clusters created based on actual behaviour.

Netflix claims that this technology is so effective that it can help to decrease cancellation rates because people stop browsing and begin watching instead. Hence, two users watching the same program will have different thumbnails and “Because you watched…” recommendations.

However, YouTube has a bit of a twist in its recommendation algorithm. While Netflix aims at showing relevant content, YouTube wants users to keep watching the videos.
Therefore, their algorithm works through:

This explains why YouTube recommendations are often sending users into a rabbit hole. The algorithm is optimized for an entire session, not individual videos. A perfect example of how the choice of metrics affects the behavior of the entire system.

Customized playlists such as Discover Weekly and Release Radar, which essentially operate in a way of a data pipeline that is scheduled weekly for every single individual user.

Spotify is an excellent example of how analytics operates on several different kinds of data simultaneously: behavioral data, text data, and raw audio signal, all integrated in one recommendation.

If we strip away all the products, the strategy of all three companies becomes clear and similar:

Constant A/B testing to ensure that the forecasting results in higher engagement. Again, because preferences evolve, and what was popular last week will soon be outdated.

All of these things are impossible to do without quality data flows, feature extraction, and constant testing which is precisely why recommendation systems are still one of the hottest skills in the field of data analytics and data science.

Each scroll, skip, and stream you generate creates a data point. Combine this with hundreds of millions of other users, and you have one of the most challenging data science problems in the world working behind the scenes in three apps we all open every single day. And the next time Netflix magically knows exactly what you want to watch, remember this is not magic. This is data analytics.