I used to sell analytics software and big data solutions for IBM many years ago. I was in California for a sales meeting with one of the large movie studios. They were interested in using IBM big data and analytic solutions to determine the next big hit movie. They wanted to use all available ticket sales data, movie genres, actor names and popularity, etc. to pick what movies they should make next. Their thinking was that a properly trained machine learning model should be able to tell them what actors, story lines, style, length, etc. would have the biggest chance for success in the future. This would drive revenue through the roof!
It was heartbreaking for me to tell them that no ML model at that time, could predict the future preferences of moviegoers by looking at data from events in the past. It would only make suggestions based on prior film successes and failures. Once they understood, we joked with them that they almost spent $1M to learn that every movie should have Brad and Angelina in it, and be about vampires. 