Consumer basket analysis methods

“What do our customers buy?” The answer to this question allows you to take a step (and sometimes a triple jump) in the direction of customer focus and business growth. Let us analyze this issue as correctly and effectively as possible with the help of numbers. I’ll just make a reservation that we are analyzing our own database analytics. That is, we have a product line, customers and a transaction base, which reflects which customer bought what. Additional data on customers (social data, questioning), products (classifier, attributes) and orders (time, money, costs) will only be beneficial. We will move towards narrowing the audience, targeting.

Shopping structure

This is the simplest analysis. Fill in the table and draw a chart with shares of transactions by product or product category. In practice, the laborious task is the classification of goods. It is also an important question, what do we consider: share of clients, share of transactions or share of money.

User group profile

Now we want to determine the food preferences of the selected group of clients (men / women, visitors on Monday morning, etc.). We can look again at the structure of purchases, but it is much more interesting to see the difference in purchases of this group. To do this, we build a profile: the structure of purchases for the entire base (general population) and the group under study (sample). And we compare these structures. For clarity, you can enter Indicator - the share of the product for a subgroup subgroup divided by the share of the same product in the purchases of all customers. If the indicator is substantially greater than 1, then this is a characteristic product for the subgroup and vice versa. if we don’t touch on the question of representativeness, then only fractions are used from mathematics. In the picture it’s obvious that the group under study is characterized by the 2nd product and products 1 and 5 are not characteristic.

A profile can also be built both in terms of the number of customers and the number of transactions (orders) or money.
')

Recommendation system

We narrow the audience to one client and create a system of personal recommendations. That is, we note (of course, analytically) that the purchase of product A entails the purchase of B with a high probability. A joint purchase of C and D (although such a combination is not often encountered) entails the purchase of E. With a sufficient number of such rules, with the proper automatic construction and sequencing, we get recommendations for each individual, but more on that in further posts.

This article is in my blog.