...

The purchase history should clearly play a role to define what other products a customer might be interested to buy (of or what products he might want to buy again).

However, a customer who bought a type of product twice or three times doesn’t have the same affinity towards these products if he bought twice or three times in total, or if he bought already 20 times or more. The first gives (so far) a clear indication of a strong affinity while the second might have stronger affinities to other products.

To make the matter worsemore complicated, some types of products have very difference purchase frequencies than others. For example, chewing toys are typically bought much less often than dog food. Therefore, simply using order frequency doesn’t suffice either.

...

The goal of calculating what we call Product Segment Affinities is to identify products individually for each of your customers where a reorder is really likely or which they buy so often enough that they might be interested in other best-selling products in the same product segment which they never bought before.

What is it?

The Collaborative Filtering on Purchases Personalized Product Affinities [Reorder | Discover] identifies what products each customer is most likely to [Buy again | Buy for the first time]. It generates a small quantity (typically around 10between 5 and10) product recommendations which are generated for each customer individually based on their purchase history.

The best practice matches other customers with similar purchases and make recommendations based on the patterns of what is often bought by these matching customerscompares the purchase frequency of each customer on product segments and compares these frequency with the average purchase frequency of other customers. In each segment the algorithm either select [the most relevant products the customer could buy again | the best-selling products the customer didn’t buy yet, but is likely to buy due to the high affinity to their product segment].

...