Hello, everybody. In this lecture, we will discuss Sell Analytics, specifically, how analytics can be applied to selling, such as demand planning, product selection, pricing and promotion. I will focus on stories, pain points, solution ideas, without getting too much into the technical details. Let's first define the metrics and objectives for selling operations, then we'll talk about some typical problems. The key performance indicators for selling include revenue, customer satisfaction, the cost. The objective is to increase revenue and reduce cost. Typical problems include demand planning, product selection, pricing and promotion, and so on. One important challenge in selling is demand planning for new products. The pain points are that at a new product launch, demand is hardly predictable. Forecast errors can be as high as 200 percent, thus it is hard to plan for capacity, supply and workforce. It is also impossible to make wise decisions on product features, distribution channels, advertising budget, and so on. Analytics, in this case, can help to make more accurate forecast for demand, and assess the impact of various factors on demand. One interesting example is the prediction of the box office revenue of new movies, which is an extremely important capability for the Hollywood. Each year, statistically, half of the movies released cannot break even. The problem is that you don't know which half. As a producer, you may also want to know what kind of movie will sell, when to release, and how much money to invest, so as to maximize the profit from the movie. So, here are the top winners and losers during year 2000 and 2001. We measure the profit by gross box office revenue, less production budget. For example, Shrek was the biggest winner and made more than $200 million, while Town & Country was the biggest loser and lost $83 million. The question is, is there a way to know this in advance? Meaning, before the release of the movie or even before production. The answer to this question lies in analytics. As an example, let's look at Avatar, released in December 2009. Before its release, even the experts cannot agree among themselves, in terms of the box office of this movie. Some film critics look down on this movie due to its dark blue color. Some other box office analysts made actually quite optimistic forecast. I'm not an expert in the movie business, by using box office data for movies released 10 years prior to Avatar, my forecast for the first week of Avatar is $164 million. The actual number is 137 million. I'm about 20 percent off. So, to make a good forecast, you may not need to be an expert in the movie business, but you must be an expert in data analytics. Many variables may have an impact on the box office of a movie, such as production budget, advertising budget, choice of distributors, number of screens, release schedule, movie critics, and movie attributes such as genre and rating. You need a sophisticated mathematical model to take the impact of all these factors into account. Taking these factors into account, we trained a prediction model using historical data. Here's the graph that compares the predicted box office versus the actual box office. The red line is the 45-degree line. If all the dots, meaning movies, are on the red line, then the model is 100 percent accurate. Of course, this never happens in practice. We can see that the model works well for the movies in the red circle, but the movies in the blue circle, the blockbuster movies with high production budget seem to behave differently, indicating a different product type and so we need another model. Combining these models, we can achieve a forecasting accuracy of about 70 percent. That means, on average, the forecast error is about 30 percent. The models also identified two product types, the blockbuster movies such as Titanic with a budget of $200 million, and less budgeted movies such as Lost in Translation, with a budget of four million dollars. We find that these two types behave quite differently in some aspects, but similarly in others. Let's look at the difference first. For blockbuster movies, season, that is the release schedule, does not affect the box office revenue significantly, but rating and budget do. For less budgeted movies, season matters a lot, especially summer, but rating and budget do not matter much. What is in common is that the number of critics and number of screens have a significant impact on both product types. Even though screens and the critics impact first week box office for both product types significantly, the impact has a different dollar value. For blockbuster movies, every critique increases the box office by about $175,000 on average. Every screen, by nearly $10,000, and for every dollar of investment, the return from the first week box office is $0.41. For less budgeted movie, the dollar value is much smaller. For example, every critique increases the box office by about $36,000. Every screen by $6000. But, if the movie is released in the summer, then the first week box office will increase by $4.6 million on average, relative to a release schedule in spring and fall. As we can see, analytics not only provides more accurate forecast, but also important insights on the go-to-market strategy for different types of products.
