Please forgive me; I am here reposting a bit on Need Anticipation:
An advantage often enjoyed by e-commerce ventures not available to other types of businesses is the use of algorithms to solve technical problems too complex for humans. Historically, supply chain management, enhanced pricing discrimination, and data mining have all bolstered the income and profitability of e-commerce vendors. One promising new technology which retailers are currently investigating is consumer need anticipation – accurately predicting the desire for a good or service, and offering it at an opportune point in time. Existing data about the customer (or other customers) is leveraged to make an informed decision about what products to offer.
While marketing has long been concerned with creating the need for a product, it is important to make note of the new distinction here: consumer needs have, at the point of time of suggestion, already been created or at least lay dormant – need anticipation intends to identify the consumer’s existing willingness to purchase an item and consequently offer the item. It is also important to note the possibility of this technology to build customer closeness in a way concordant to branding by allowing the customer to spend additional time browsing recommendations, allowing users to recommend products to friends, and allowing users to rate items they are interested in. These features not only keep customers on a website for longer, they give the user solid reason not to switch to a competing website (and lose the use of their ratings) and they give the user a motivation to invite their friends (to gain and give recommendations).
Amazon may well be the most successful early adopter of consumer need anticipation tactics. When a new user arrives at the site, they are greeted with the message “Hello. Sign in to get personalized recommendations. New customer? Start here.” , along with generic recommendations of bestsellers, sale items, and seasonal items. When a user logs in, they are presented with new recommendations based on recently added products. Throughout the process of viewing products (a procedure roughly corresponding to browsing), the user is shown a histogram outlining what products were bought by users who viewed the current product, instantly showing the path of least resistance for most consumers. This entrusts that the customer can easily find the most ‘purchasable’ similar products. Suggestions are based on items “Frequently Bought With Items in Your Cart”, “Customers Who Bought This Item Also Bought”, “Frequently Bought Together” [with the currently viewed item], and “These recommendations are based on items you own and more.”  Clearly, Amazon engineers are making use of several different easily obtained statistics to make recommendations, and it seems as though most data points are being put to use in some way or another.
Technically, the statistical approach to Amazon’s approach seems to be ad-hoc and relies on the common sense translation of user data into predictive sets. Amazon does not disclose any of its algorithms, and seems mostly intent on providing for its users raw data of the type “You may like the following…” While the simple correlation of two or more items may not find a granular, specific likelihood of a user’s desire for a book, Amazon’s methods seem effective enough to bolster sales by offering the user a product they are likely to buy based on the purchasing patterns of similar users. Though other companies have pursued the same approach, Amazon’s engineers seem to have taken the lead in accurately predicting a user’s habits. 
Despite the success of this approach, more incisive algorithms are under development which attempt to predict a quantitative rating of a product based on prior ratings of other products: enter the Netflix Challenge. Netflix has built an empire on the ability to deliver their physical product (DVD rentals) quickly, efficiently, and without hassle to the consumer, but the company is less than content to rest on their laurels. By providing anonymized data to software developers, Netflix hopes to arrive at a superior ratings prediction algorithm. Basically, entrants to the Challenge are given 100 million DVD ratings entered by 480 thousand users, and are asked to generate as accurately as possible ratings for more movies. The company justifies the $100 million US prize on the basis that the basis that “Netflix is all about connecting people to the movies they love.” It is clear that Netflix is forward-looking in their desire to suggest to users movies that they will like (thereby increasing use of their service), and admirable that they are advancing the state of the art in order to increase profits. Users have so far created entries based on expert systems, neural networks, and other advanced artificial intelligence algorithms.  The Challenge itself is an innovative form out outsourcing that may be saving the company money: According to Vivek Ranadive, “Holding on to a competitive edge means staying one step ahead, and the more reliably one can predict the next step is often the difference between success and failure.”
1 Wikipedia: Amazon
5 Ranadive, V. (1991). The Power to Predict. McGraw-Hill (January 2006)