Association Rules
Some people call them “Related Selling” analyses. Others call them “Market Basket” analyses. Regardless of the name, association rules are used to examine the relationships that exist between the products you sell and the customers that buy them. These rules are typically expressed as probabilities, percents and likelihoods. For example, 10% of my customers buy Jewelry and Handbags, 50% of the customers who buy Denim also buy Polos, or Denim customers are 3.5 times more likely to buy Sunglasses than any of my other customers. The goal of the analysis is identify the co-occurrences of different products that happen with the greatest frequency. By performing this analysis you will be able to answer the questions posed above.
CustomerTool's association rules application analyzes POS transaction data to find the associations between the products your customers purchase. You do not need to be a statistician to use CutomerTool. All you need is access to the data. There is no hardware or special software to buy. Just upload your data and start getting answers.
Association Rules: An Example
CustomerTool produces 4 association rules: Expected Confidence, Support, Confidence and Lift. Each of these rules has a specific use in evaluating selling data. See the example below for a quick description of these rules.
Expected Confidence Is the probability that any product you sell will be purchased by a customer. For example, a store had 100 customers today, and 20 of them purchased at least one shirt. Consequently, there is a 20% chance (EC) that any customer transaction will include a shirt.
Support Is the probability that any combination of products you sell will be purchased by a customer. If 10 of the 20 customers who bought a shirt also bought a hat, then shirts and hats have a 10% chance (Support) of occurring together on any transaction.
Confidence Is the probability that if a customer buys a product(s) they will also buy another. Using our example above, shirt customers have a 50% chance of also buying a hat (10/20).
Lift Is expressed as a ratio and measures the likelihood of buying one product given that another product has been purchased. Lift is calculated by dividing Confidence by Expected Confidence. In our example, someone who buys a shirt is 2.5 times more likely to buy a hat than someone just walking in the store. NOTE: It is possible to uncover lifts less than 1. This means that the products are negatively correlated and actually make someone less likely to buy another product.
|
A graphical example
|
Links to resources on Association Rules
| 