Select Page
November, 2017


As viewers, movie trailers tell us all about a movie so that we can decide if we’re interested in seeing it or not. For the studios, trailers are a key promotional tool exactly for this reason. For the data scientist, trailer interest is a leading indicator of box office success … or so the theory goes. This is a hypothesis we’ll explore today.

First it is important to state our hypothesis clearly: more trailer views = more box office revenue. This seems intuitively correct, but specifics matter, and so we need to test it.

Our immediate challenge is, of course, source data. Sadly, there does not exist a master listing of trailers and views that we can simply reference. Since FLIQ has been online, it has been scouring the web for trailers, and right now it has an average of 9 trailers for every movie in the database (and it is being selective!). Trailers are not only released by the studio. There is an entire cottage industry of companies specializing in releasing trailers, and even fans will make trailers that in quality and viewership compete with official releases. This is clearly more complex than just looking up views for a single clip.

Moreover, the idea of a trailer itself is flexible. There are trailers, teasers, clips, behind the scenes, featurettes, convention videos, etc. All this content promoting the movie is being released pretty much all the time, at irregular intervals. It’s all competing for attention, and has the potential to tell us something about how many people will walk into a theater and buy a ticket for a specific movie many months from now.

We’re going focus on Thor: Ragnarok in this analysis. It’s coming out this week, so this analysis is immediately useful if anyone is making their own prediction, and we can see what actually happens at the box office this weekend. Consider the “official” trailer above. This video was released by Marvel in June and has 47 million views, which is valuable information (and a lot of views!). We should be able to compare these views to the views for other Marvel movies, and get an idea of how much money Thor: Ragnarok will make based on what the other movies did.

This is where the complexity of multiple videos kicks in however. Marvel also released another trailer for Thor back in April, and that one generated 44 million views, so it was also big. What should we do with that one? In fact, when we look around, there are 11 clips of one sort or another for Thor that all generated significant views. Should we just combine all of them?

Even if we do that, we run into the issue of timing. We would be comparing these trailers with ones that came out years ago. The top trailer for Avengers: Age of Ultron has 81 million views, but it was published back in 2014 and has been collecting views for the last 3 years, which does not make for a good comparison.

Normalized Metrics

What we really need to do is normalize these views so that they are useful to us for comparison. This is something that FLIQ does. It collects all available trailers for a given movie, maps their viewership over time, weeds out the riff-raff and leaves us with key trailers which are responsible for at least 80% of all views for a given film.

This is data that essentially allows us to answer the question,  “How many views did the promotional videos for this film generate in their first two weeks of release?” and compare the results to other films.

In the chart on the right, we took all the movies from the Marvel Cinematic Universe and compared them using this metric. Is this information useful for predicting how much money a movie will make? We added a column for opening week revenue to help us understand this. Just visually, it’s clear that there is a relationship between these views and revenue. Essentially, the more views the trailer videos generate, the greater the revenue.

Analysis above excludes Iron Man I & II, Thor, The Incredible Hulk and Captain America: First Avenger due to data sparseness.
Data by FLIQ.AI

It’s worth noting that it’s also clear visually that the relationship between these views and revenue is complex and evolving rather than linear, and these numbers can be refined further – but they are helpful already.

Marvel Cinematic Universe: Upcoming Releases

One of the big things we were interested in with this analysis is how the upcoming Marvel movies stack up. We can see in the chart that Thor: Ragnarok is performing really well, second in views only to Age of Ultron and Civil War, which signals it should be one of the highest openings of the Marvel releases. Black Panther views are also strong, but not as high, and are closer to this summer’s Spider-Man: Homecoming, which would still make it one of the best performers at the box office. Further out we have Infinity War. I believe the low views here are more indicative of the fact that it appears their trailer campaign hasn’t yet started in earnest.

For overall context: the top performing movie that we have seen using this scale is Star Wars: The Force Awakens with 121 million combined views in the first two weeks. The average major release over the last 10 years generates about 8 million views, so these films are all at the upper end of the performance spectrum.

Thor: Ragnarok

Taking a deeper look into Thor and this data is revealing. In the following chart we show cumulative views across all trailers (top), and a count of videos released (bottom). Part of the story here is just how much more active the trailer space is in 2017 than it was in 2011, but the real thing to note for our purposes is that initial growth ramp that aligns with major trailer releases. This is exactly the burst FLIQ tries to capture with the metric we described above and why it acts well as a leading indicator.


In addition to views in the first two weeks, the overall shape of the view curve matters and so does the number of trailers released. It is clear that there is real useful information in trailer data.

For the FLIQ predictive models we use trailer activity combined with search trends and Wikipedia activity data to understand the Audience Interest elements of an upcoming release. We’ll be delving deeper into this topic in the future and cover some technical aspects of capturing and using this data to understand and forecast movies with machine learning.

We’ll also be posting our final forecast for the upcoming Thor this Friday, so come back for that.

Sign up for the newsletter below! Thoughts/questions welcome =)


Want new articles before they get published?


@2017 FLIQ.AI - All Rights Reserved