It’s no surprise that current industry trends have raised the stakes for content companies to know and cater to their audiences. The power of the connected consumer, digital influence in a social world and the ubiquitous distribution of content on mobile devices are among the factors requiring that movie studios increasingly focus on delivering audience intelligence capabilities to better enable the media enterprise business teams.
With the use of big data analytics, previously siloed data sources from inside and outside of the media organization can finally be “unified and utilized” for game-changing applications such as demand forecasting, fan engagement scoring, marketing campaign effectiveness and audience and segmentation. Through more accurate understanding of audience demand, business teams can determine far in advance when, where, why and how particular actions need to be taken.
But still, the question remains: How do media companies predictively evaluate demand for their content or services? The answer can be broken down into four key areas:
- Identify measureable target outcomes and KPIs
- Determine audience behavioral proxies
- Build predictive models and demand scoring matrices
- Integrate predictions with business decisions
As we consider what these areas mean in terms of real-world use cases, we can ask a series of questions that no doubt ring true for every studio exec. For example, we can consider forecasting retail demand for packaged media and ask “How much should I sell-in to retailers to optimize sales?” Or we can analyze content service churn and ask “When should I take action to prevent subscriber loss?” Demand scoring for content archives begs the question “what content should I digitize and clear for licensing?”
Finally, predicting a movie’s Opening Weekend Box Office or OWBO, the focus of this post, leads us to inquire “How do I know when to dial-up my marketing?”
Answer that question and you’ve got a Holy Grail in your hands.
Movie marketing’s most critical KPI is OWBO, but until now marketers had yet to find an approach to correlate audience behavior with box office outcome. Seeing as the movie marketing timeline starts at roughly 12 weeks out and tends to ramp up 2 to 4 weeks out from opening, there is tremendous potential for over- or under-spend on multi-channel marketing campaigns that typically range in the tens of millions of dollars, if not considerably more.
IBM recently engaged with a major movie studio to build a box office prediction model based on online audience behaviors. Our goal was to determine if there was a predictive relationship between social data and OWBO. We also needed to determine which variables are the strongest predictors, how accurately we can forecast box office performance, and what types of movies have a higher or lower forecast accuracy.
We determined predictive power by collecting a wide variety of data including, but not limited to, movie characteristics (movie size, genre, studio, seasonality, rating, etc.) and online presence (Twitter volume & sentiment, Facebook Likes, Shares and “People Talking About This,” and other Internet indicators such as trailers, reviews, press, blogs and more.) We then trained models based on historical data from over 200 movies encompassing a wide variety of genres and audience types. At that point, we could evaluate models for accuracy at times that directly correspond to the average movie marketing timeline. We found that there are clear and present relationships between social signals and box office sales; in particular, Twitter volume and negative sentiment have a strong correlation with actual OWBO results.
As such, our approach resulted in the highest prediction accuracy versus current industry benchmarks.
In fact, we achieved high degrees of forecast accuracy up to 8 weeks out – a timeframe in which marketing campaigns can still be changed. That can make a would of difference between a blockbuster OWBO and a need to quickly shift funds from another movie’s marketing budget or perhaps another geo-location you now know won’t perform as well as you initially “thought” it might.
Although we’re talking about movies and movie studios in this example, you can see how these types of predictive applications of big data analytics in media and entertainment could be used to predict performance for almost any type of content, campaign or program.
In my next post, we will explore additional details of our movie marketing prediction models and share some more information about what we learned. I’ll also share some particulars about how the unique characteristics of genres like horror and animation can help us look at typical social data in dramatically atypical ways.
And I might even share some thoughts on Ben Affleck being cast as the next Batman.
Read the sequel to this post: "Predicting Relationships between Social Signals and Box Office Sales"