Client: A fashion retailer

Project: Create market basket application for merchants

Challenge: Extend the output of a standard association algorithm to include KPIs such as Units per Transaction, Average Order Size and Total Basket Revenue. A previous vendor had attempted the project but had failed to scale the application.

Achievements: While many machine-learning researchers have improved the speed and scalability of association algorithms, business users want more than counts and conditional probabilities. The classic “beer & diapers” analysis is of limited value, but when marketers can see the statistics they need for each basket, they are in a position to interpret the analytical output and make business decisions with it.

One team of consultants previously attempted to develop the application with a BI tool on top of a relational database. Our solution consisted of leveraging the classic a priori algorithm to create a design matrix of combinations that a given product had with other members of the merchandise hierarchy. That enabled the calculation of additional statistics to take place in an efficient way.