Introduction

Market basket analysis involves the use of data mining techniques to search for sales patterns between products within a given group of transactions. The output of that analysis provides a rule that defines the association found between products at the subclass or class level of the merchandise hierarchy.

A rule consists of one to three antecedents (IF attributes) and a single consequent (THEN attribute). For example:

IF (milk) and (juice), THEN (cereal)

In other words, if a customer purchases an item from the subclasses milk and juice, the customer will also purchase an item from subclass cereal. After a rule is defined, a user can use the AA interface to understand how strong the affinity is, using rule confidence and support. The probability that a customer will buy milk, juice, and cereal is known as the support percentage, while the conditional probability that they will buy cereal when they buy milk and juice is known as confidence. Rules with a very high support value occur frequently in your transaction history, while rules with a high confidence value represent a strong affinity between products.

After users have identified selling patterns, they can begin to take action based on those patterns, as well as the needs and goals of their product category. Suppose that a merchant is tasked with bringing in more margin dollars to the cereal category. Using the affinity rule in the preceding example, the merchant might work with the dairy category on a milk promotion to increase sales of milk. This in turn increases the sales of cereal, without sacrificing margin dollars on a cereal promotion. Note that this can require cross-category planning in some cases, as product affinities can sometimes occur between seemingly unrelated products (such as pet food and beer).

Another component of market basket analysis relates to the product assortments being sold in stores. Using product affinities and sales history, AA provides assortment recommendations that improve the revenue or margin of a category by suggesting product additions or removals. Products may be recommended for removal if they are found to be too similar to other products in the assortment (and thus cannibalize the sales of those products). Conversely, products that are not similar to any other items in the assortment may be candidates for inclusion, as they will not divert sales from the existing assortment.

Market basket rules are used to improve the assortment recommendation by showing the potential lift (or halo effect) on your overall sales due to any known affinities on recommended item additions. For example, if AA is analyzing an assortment for Coffee, and a particular item is part of a market basket rule that drive sales for Milk, then that item has a greater potential value for the lift it brings to the Milk category. AA may then recommend that item over other items in the category, because including it will bring in additional revenue to other assortments without changing those assortments directly.