The honest answer is rarely the algorithm.
Cross-selling is the practice of offering an additional, complementary product to an existing customer. Different from upselling, which sells a higher-tier version of the same product. The strategy is decades old. Most companies still do it badly.
In 2003, Greg Linden, Brent Smith and Jeremy York published "Amazon.com Recommendations: Item-to-Item Collaborative Filtering" in IEEE Internet Computing. The paper described a method that scaled to tens of millions of products. In 2017, the journal awarded it the "test of time" prize.
That paper now lives on the homepage of every retailer in the world.
Per McKinsey, Amazon's recommendation engine drives roughly 35% of total sales. On a $470 billion revenue base, that is over $160 billion of cross-sell.
For most other companies, the same idea collapses.
Gartner research finds that 74% of companies try cross-selling. Up to 90% fail. McKinsey found that only 15% of customers are actually receptive to cross-sell offers. CEB found 85% ignore any offer not relevant to what they already need.
A 2012 Harvard Business Review study went further. One in five cross-buyers was unprofitable. Those customers accounted for 70% of the merchant's total shortfall under the strategy.
The pattern repeats in every failed example.
The retailer treats cross-selling as an extraction tool. Add this to the basket. Buy more, faster. Hit the quarterly number.
Amazon treats it as a discovery service. The "Customers who bought this also bought" line is computed based on differential probability, not on bestseller count. The recommendation has to be more useful than what the customer would have found on their own.
That distinction is the entire game.
Cross-selling is not a feature you add to checkout. It is a question you answer for the customer: what else will genuinely help them?
If the answer was constructed to hit your number, the customer can already tell.
Where in your last cross-sell campaign did the metric come before the customer?
