Administer > Manage WebSphere Commerce features > Marketing > Collaborative filtering > WebSphere Commerce Recommendation Engine powered by LikeMinds


Example: LikeMinds usage

The following examples illustrate both appropriate and inappropriate situations for deploying LikeMinds:


Business-to-consumer clothing retailer

A good example of using LikeMinds to recommend items is with an online clothing retailer. Successful business-to-consumer retailers typically measure up well on the following three key LikeMinds evaluation metrics:

Suppose a retailer has about 500 inventory item IDs ranging from jackets to pants to accessories, such as make-up, belts, and hats. The retailer's Web site has a return visitor rate of once every two months, and an average shopping cart checkout of 3-7 items per visit. While shopping the site, the average user may additionally perform product detail views of another 10-15 items. This user may add another 2-5 items to his or her shopping cart, remove 2-5 items from the shopping cart, view a couple of articles on the future of clothing design, sign up for a mailing list, and so on. In total, a typical shopping experience at this site may include as many as 30-40 user-to-item interactions of varying importance.

In this case, we can assume this retailer has an inventory of about 500 item IDs and that there are about 150,000 visits or shopping sessions per month, with a conservative estimate of 15 user-to-item interactions per session. This results in about 2.25 million user-to-item interactions per month. Averaged across the total number of item IDs (which is not a very accurate methodology), that results in a very rough estimate of 4500 user-to-item interactions per item per month. Even with seasonal (biyearly) turnover of inventory, this is considered good user-to-item transaction across item inventory coverage.

This body of data is a rich source of information about both the user behaviors and the item relationships. This example would be an excellent fit.


Credit card vendor

The following is a bad example of deploying LikeMinds. Suppose the potential customer is a credit card company that has 5-20 different credit cards offered. In this case a typical customer interaction might include a user looking at all the cards (product detail views), comparing and reviewing their benefits and features, and then signing up for one credit card.

In such a case, the only truly significant transaction that has occurred is the user finally signing up for the one credit card. With this example, there is only one user transaction and one user-to-item transaction per shopping session. Additionally, there is very little return business; a customer signs up for credit cards only once every 1-10 years. Due to very low return visit rates, there is a very low user-to-item transaction coverage across the item inventory.

The question to ask yourself about this or any LikeMinds usage is, "Can a community based recommendation be derived from such user-to-item transaction data?" The answer in this case is that it cannot. In this situation, recommendations would be best derived from a rules-based system. This is because credit cards are not typically promoted or chosen based on a person's opinion. This type of promotion is usually done based on business rules. For example, customers with better credit typically qualify for cards with higher balance limits and lower annual percentage rates, and so on.

In short, the credit card example is probably a bad environment in which to deploy LikeMinds. The reasons are...


Wedding dress retailer

In general, a wedding dress is something used once or at the most a few times in a person's lifetime. The transaction data that is gathered from people shopping for a wedding dress, at a retailer selling only wedding dresses, would typically be considered insufficient to build a powerful enough relationship between users and products.


+

Search Tips   |   Advanced Search