Data Sheets

Affinity Model Quick Guide

Issue link:

Contents of this Issue


Page 0 of 0

Predictive Product A!nities Overview S A L E S @ C U S T O R A . C O M Methodology Custora starts by observing each customer's unique purchase history. We record the order size
 of each customer for each item. To avoid outliers, a!nities are calculated using items that make up the top 95% of sales. Because customers do not explicitly
 rate items, we use a customer's purchase history as implicit feedback. Use Cases Niche Product Targeting To market a niche product without replacing your daily email, send the product-specific email to the recommended a!nity segment and the daily email to everyone else. Dynamic Email Personalization Create a curated customer experience by using dynamic content capabilities to populate email content based on each customer's predictive product preferences. "Has Ever" Audience Expansion Identify a larger audience for product-specific campaigns and target all customers who have an a!nity for a product type instead of just customers who have purchased it in the past. Product Markdowns Leverage a!nity audiences to be more targeted when sending product markdown emails, targeting only those customers that are most likely to be interested in the given product. Custora's a!nities are determined by applying the Alternating Least Squares (ALS) model for implicit feedback. This model is commonly used to estimate preferences and provide recommendations. Custora's a!nity model identifies the customers who are most likely to purchase a specific brand, category, or sub-category. In other words: it identifies the right audience for a given product. Then, Custora uses a collaborative filtering technique called matrix factorization and employs Alternating Least Squares (ALS) as the method for optimization. Matrix factorization can be thought of as
 a lookalike model: it predicts customers' behaviors based on the observed tendencies of similar customers. This allows Custora to predict a!nity scores. These are pairwise scores between each customer and each item in the dataset. Last, Custora recommends an audience size for each a!nity segment based on what has performed best historically. It is the optimal trade-off between size and
 the likelihood that
 the customers in
 the audience will be interested in the item. 1. 2. 3. 4. For advanced detail on how our models work, we recommend this article on Collaborative Filtering for Implicit Feedback Datasets: Size Interest Q U I C K G U I D E

Articles in this issue

Links on this page

view archives of Data Sheets - Affinity Model Quick Guide