Imagine you’re a retailer and you’re trying to plan your next line of products. What information do you need to know? A useful way to look at it is by exploring attributes—the variables of the product and the customer base. Do wealthy suburban women prefer blue or green purses, and do they like them to be traditional or fashion-forward? Which purses do you already carry in blue or green, and just what the heck is meant by traditional, fashion-forward, and everything in between?
The value of attribute analysis expands across industries. An entertainment company—say an HBO or a Netflix—needs to know what current movies and TV shows its customers like so it can better decide which future movies and shows to buy or create. Do they prefer longer movies or shorter ones, happy endings or sad ones, scary or comedic or dramatic themes? If you want to know these things, it’s very useful to know the attributes of entertainment offerings.
A roadblock to use is that, in many industries, there is no widely accepted taxonomy of product attributes, and many manufacturers don’t classify their products. So companies that want to do some analytics need to create their own.
Apparel manufacturers, for example, don’t classify their products in any systematic way. So leading retailers are spending considerable time and effort classifying product attributes on their own. Zappos, an Amazon subsidiary specializing in shoes and leather goods, involves three different departments in product classification so that it can optimize customers’ searches and create the most effective offers. The classification involves product type, style, color, pattern, brand, and price. This can get complex: Customers can choose from more than 40 different material patterns—pearlized, patchwork, pebbled, pinstripes, paisley, polka dot, plaid—alone. You need to know that a customer has bought patchwork-patterned goods in the past to be comfortable recommending them in targeted offers.
In entertainment, the king of attribute analysis is Netflix. There is a decent classification system available from IMDb, but Netflix thought it could derive competitive advantage from a more detailed classification structure—with almost 80,000 categories of movie types, as well as their actors, directors, and so forth. The company uses human classifiers to do this work, and has a 36-page guideline to attribute classification.
Netflix, of course, uses the attributes for its movie recommendation engine, but it doesn’t stop there. The company has also used the attributes to predict commercial success, classifying attributes of shows before creating them. In doing so, Netflix has been able to substantially improve its success rate in developing and buying entertainment products.
For example, in the case of the very popular series House of Cards, Netflix increased the likelihood of its commercial success by classifying its likely competitors and the popularity of its actors and director. The closest competitor was a UK series with the same name. Kevin Spacey, a popular actor in Netflix shows, plays the evil president in the show. David Fincher is the producer. Netflix observed high correlations between all three attributes and commercial success.
This approach works. In addition to House of Cards, Netflix has produced many other original shows that gained loyal viewers, including Orange Is the New Black and Unbreakable Kimmy Schmidt. More than 90% of Netflix’s original shows were renewed after their first seasons—well over the recent 35% success rate of the company’s TV network competitors.
Attribute classification and analysis is now being used to predict the success of another entertainment category—the novel. A new book, The Bestseller Code, describes an algorithmic approach to identifying best-selling novels. The approach, developed by a professor and an editor, employs 2,800 attributes, including theme, style, vocabulary, and punctuation. The developers claim an 80% level of prediction accuracy.
So if your company’s promotions aren’t succeeding, or if your new product or service success rate is low, take a cue from these aggressive adopters of attribute-based analytics. Classify some of the key attributes of your past and current products or services. Figure out what customers prefer what attributes. Or analyze the past relationship between those attributes and the commercial success of the offerings. You’ll have a predictive model that should give you some sense of how likely a customer is to buy a particular offering, or how likely a new product or service is to be successful. With these types of attributes and analytics you’ll better understand your own offerings and how customers will feel about them.
Tom Davenport, the author of several best-selling management books on analytics and big data, is the President’s Distinguished Professor of Information Technology and Management at Babson College, a Fellow of the MIT Initiative on the Digital Economy, co-founder of the International Institute for Analytics, and an independent senior adviser to Deloitte Analytics. He also is a member of theData Informed Board of Advisers.
Subscribe to Data Informed for the latest information and news on big data and analytics for the enterprise.