Cross-sell recommendations are bought with the current product; thus, it is best to use sales data to determine cross-sell. Engagement, such as view data or interaction with products, is good for finding similar products, where similar products are an alternative to the viewed product. Engagement can harm cross-sell since the customer looked at, but never bought the product.
Furthermore, cross-sell should not be solely based upon the common customers, but use complex math to take into account the fact that the customer is already interested in the product, as well as coincidence of two products being bought together versus a true relationship between the products. This helps sell deeper into your catalog and not just recommend the top selling products, unless appropriate. Recommending the same top sellers decreases the customers confidence because they feel like you are trying to sell more and not help them find the products that they want.
Cross-sell is valuable when a customer is viewing a product detail page because they may buy the product and the cross-sell recommendation (or even buy the cross-sell recommendation on its own since they own a similar product to the viewed product). Cross-sell is also valuable when a customer is adding the item to the cart, checking out, confirmation emails, and cart abandonment email. We’ve found increases in sales up to 25% from adding cross-sell to only your checkout process.