Item Affinity Engine

The Item Affinity Engine generates recommendations based on any transactional history available, such as shopping cart activity, external legacy transactions, and web transaction completely unrelated to shopping cart activity (page views, product inquiries, searches, and so on).

An Item Affinity Engine can, for example, predict that a user who purchases a digital camera would likely want to purchase compact flash cards or a USB card reader. The Item Affinity Engine also spots the less obvious connections—that users who purchase beer are likely to purchase diapers at the same time, for example.

The Item Affinity engine lets you track more than mere purchases measured at check-out time. It can identify items the user only considered for purchase. For example, it can know when a user only considered purchasing rye bread rather than actually purchasing the rye bread. (This is measured by a shopping cart add, followed by a shopping cart drop, of the rye bread.) In this case, the grocer could not possibly know that the user considered purchasing rye bread during the shopping session, since the rye bread was not in the shopping cart at check-out time. Even though a considered purchase does not necessarily imply the same level of item affinity as a completed purchase, it does convey item affinity information.

Unlike the other engines, Item Affinity Engine recommendations are based on Market Basket Analysis statistics, not collaborative filtering. Market Basket Analysis enables content affinity predictions even when cold-start situations obscure the relevance of collaborative filtering. The Item Affinity Engine can be used to provide improved automated recommendations, such as cross-sells, even for first time visitors to the Web site.