Product Recommendations
Product recommendations are generated from a recommendation engine to increase repeat purchase rate and frequency, focusing on advancing first-purchase customers to repeat customers.
The recommendation engine is generated with a combination of different algorithms, including user clustering in combination with predictions of future product purchases, to produce the following outputs:
- Advanced Segmentation Clusters: user-level segmentation clusters useful for segmented targeting and A/B segment testing.
- Cross-Product Weights: Tabulation of product cross-sales is useful for answering the question, “Customers that bought x also bought y.”
- Top-n Product Recommendations: user-level product recommendations for the most likely next purchase based on product and user similarity, as well as input weights and rules.
Hierarchical clustering is used to generate segments of similar customers. It allows us to rank order how similar one customer is to all other customers. Hierarchical clustering utilizes user attributes, such as:
- Purchase Behavior
- Demographics
- Geography
- Predicted Revenue Features
Chord then adds a distance metric, for which the user attributes are fed into an agglomerative distance algorithm to minimize “close” and maximize “far” relationships.
Once these similarities are generated, we can use them to identify what other similar users have purchased.
Another view into future behavior is the relationship between what one customer has purchased and related “cross-product” purchases. Collectively known as collaborative filtering, this process builds weight from user-item relationships and uses them to predict likely future purchase behavior.
The final output is the top five product recommendations per user. These predictions are running daily and updated based on new or learned data points. These outputs can be activated in a variety of ways, such as email campaigns with personalized product content.