Prefence predicting - long-tail of curve

Recommendation engines deliver items that more closely meets customer’s wants. Traditionally, web pages display “other users like.” This is similar to offering the movies Transformers and The Hangover. These items are the popular items and lie in the fat of the bell curve. They are a small percentage. This ignores the lesser known items in the long-tail that will be enjoyed by the customer. Netflix delivers movies that are lesser known but that receive high ratings by the user. It uncovers the unknown. Otherwise, only a handful of popular movies would be watched and the customer would end his/her subscription.

Online stores increase the popularity of popular items by offering "other users like" to the buyer. A user becomes aware of the item and is more likely to purchase the item thereby making the item more popular. The effect is a self-perpetuating popularity list. New items that have not been purchased will not make the "other users like" list regardless of matching a user's preferences.

For the online store, items that meet a customer's wants are displayed. Given the limited space available for user browsing, hitting a customer's wants is more valuable.

Google leverages Predictive Analytics

Predictive Analytics

Google has applied the methods successfully across its enterprise. Netflix keeps users' queues packed through it. Pandora is reinventing the radio with it.

Before Google, internet search was sorting through 1,000 items. Google simplified internet search to the 3 most likely to match a user's need. Google beat Dogpile and Yahoo by applying predictive analytics.

Recommendation engines will improve online stores by turning searching into finding and in return improve the company's profit.

Don't Need to be 100% Right Just Predictably Right

Predictive analytics are a prediction of a customer’s actions. They are an improvement over present tools.

Similarly, ERM and CRM systems do not create 100% correct forecasts. They are an improvement over the tools prior to these systems. They improve the decision and thereby improve profits.

A recommendation engine creates a more accurate prediction and leads to better decisions. The engine increases the chances of targeted communication and matching customer’s needs over a non-predictive based system. “Predictably right” lowers costs of communication and improves revenue by more closely matching the customer’s actions.

Predict Customer Actions

The better a system can determine the customer’s actions, the better the company can target the customer. A recommendation engine can predict the customer’s actions and increase profits.

For example, a company can lower its costs of customer acquisition by knowing which customers will become repeat customers through the offer of a coupon. The system can determine the 60% of customers who will become repeat customers. The targeting of this set rather than all customers lowers the costs and improves revenue on the 40% of customers who are inclined to become repeat customers without a coupon offer.

Simplify View of Products

A recommendation engine can lead to showing a targeted set of items, focusing the customer on the more desirable items. The engine driven by predictive analytics determines the items that the customer is more likely to enjoy. This improves the decision process by removing information overload.

For example, Google Search returns the most likely results first. Google’s predictive analytics removes the need to click through pages of insignificant results. Similarly, Amazon shows items that likely fit the need of the customer, allowing for ease of viewing and decision making.

Mike Shor, a professor at the Owen Graduate School of Management, wrote an article on the benefits of simplifying the decision-making process by decreasing the number of items offered.

Excerpt:

“...often-cited experiment in which people are given the chance to taste a sampling of jams, asked to pick their favorite, and given coupons to buy one. One group selects from 6 jams, another from 30. Researchers found that people who tasted 6 were more likely to buy something than those who tasted 30—the latter simply had too many choices.” - http://owen.vanderbilt.edu/vanderbilt/About/faculty-research/featured-research/effects-of-complexity-on-decision-making.cfm

Drawing out Tastes

Consumers buy based on their tastes. Their tastes have patterns that predictive analytics can model. As a customer gives feedback on items through a rating system such as Netflix's the patterns describe the customer’s buying preference.

The patterns may be logical, such as enjoying a particular director or agreeing with a certain critic. However, the patterns may be more difficult to decipher by human effort. Predictive analytics proves these patterns, both the logical and the not obvious, removing the need for human intervention.

The Netflix recommendation system through the rating patterns determines the user’s preferences for movies. This system introduces movies that meet the user’s tastes, increasing the satisfaction of the Netflix service. As well, the system can recommend lesser known movies that match the user’s tastes, increasing capacity use of all movies.

As well, the Netflix system removes items that I am less likely to enjoy. Netflix removes the noise leaving items that have greater satisfaction.