Market Basket Analysis is a technique that shows you which products are sold together with other products for reasons other coincidence or their independent popularity. It can show you which products you should consider promoting in order to increase the sales of other products. Watch this short 10-minute video by Rafal to get an introduction to this very useful analytical data mining technique, and to start gaining a better understanding of your sales patterns, and also to see what the results of such analysis could look like. Log-in or get a free account to watch it!
In essence, basket analysis uses associative analysis, also known as the Association Rules algorithm. It first finds the co-occurring itemsets, and then the rules, that focus on those products that have a non-trivial association with the other products with which sell together. You can analyse such connections from a perspective of a single visit to a store, or you can take a longer-term approach to what makes up a basket, and even consider a customer’s lifetime purchases from your company, such as in the case of a pricier, luxury-goods retailer, or perhaps a car dealership.
A common mistake, however, when analysing carts or baskets, is to just find a grouping of items that sell together. This is not useful, because those items might sell together simply because they are—independently!—popular. Imagine a store which only sells 2 unrelated goods, say milk and glue. Assuming for a moment that running such an odd retail operation would make sense to someone, a junior analyst looking at the sales of this outlet might assume that since milk is often sold with glue, there must be a meaningful association between them! In some real-world scenarios we have seen real customers promote such truly unrelated items because those showed up in an analysis as having sold together. Needless to say, the results of such promotions were disappointing.
Of course, making recommendations based on the fact that some items just sell together often enough does not make sense. Instead, we ought to look at the so-called rule importance of the association between the items sold together, also referred to as the interestingness score, which takes account of the likelihood of selling one item not just with the other one, but also, more importantly, without the other one. If you would like to understand that, make sure to watch the in-depth module on Association Rules, which explains, in detail, those metrics of the strength of association, and which will help you build great recommendations and promotions.
