Methodology Behind Product Recommendations
The recommendations are based on frequency and similarity, which are divided into three levels: customer specific, product specific, and customer segment specific.
LEARN MORE: Check out the complete set of methodologies and corresponding examples, here.
How To: Learn to define the product recommendation model, here.
How To: Learn to configure the product recommendation model, here.
How To: Learn to revise the product recommendation model, here.
Customer Specific Recommendations
Customer-specific recommendations, such as "Frequently Purchased," are calculated based on how often a customer bought a specific product. The frequency rate is determined using the count of baskets of the specific customer as the denominator and the basket count of customer-product purchases as the numerator. This frequency rate is then multiplied by the product's historical profitability metric to generate a recommendation score.
Product Specific Recommendations
Product-specific recommendations, like "Bought Together," are based on how often two products were purchased together. The frequency rate is calculated using the count of baskets with the quoted product as the denominator and the basket count with product and co-product purchases as the numerator. The recommendation score is determined by multiplying the frequency rate by the co-product's historical profitability metric.
Customer Segment Specific Recommendations
For customer segment-specific recommendations, such as "Others Buy," the frequency rate is calculated using the count of baskets of the customer segment as the denominator and the basket count of customer segment - product purchases as the numerator. The frequency rate is then multiplied by the product's historical profitability metric to generate a recommendation score. These outputs are stored in the Model Table "Customer Segment Recommendations."
How To Generate Recommendations for Similar Products
There are several methodologies for generating recommendations for similar products, similar products from a specific brand, up-sell products, and down-sell products. The outputs from these are stored in respective model tables for further analysis and implementation.
In addition to transactional data, product-to-product similarity is also considered in generating these recommendations.
LEARN MORE: Check out detailed calculations for each method and explanations how the recommendations are ranked and stored, here.
Good to know: If no customer segment is present in the transaction data, users can generate a customer segment based on product affinity.