Challenge

With its huge customer base, the retailer struggled to:

• Provide personalized product recommendations
• Measure the impact of personal discounts
• Manage the product matrix though targeted offers to customers interested in specific products

Project objectives

The goal was to boost key performance metrics:

• Increase margin by 1-2 p.p.
• Raise the average number of purchases by 20%–30%
• Expand the active product matrix by 10%

Solution

We developed a hybrid recommendation model combining neural network embeddings and collaborative filtering, designed to:

• Leverage personal purchase histories and customer preferences, even with very large customer bases
• Ensure continuous retraining on various user signals, such as purchases, searches, comparisons, cart activity, etc.
• Predict future purchases based on purchase frequency