Shopify's generative recommender predicts which products to suggest from the sequence of a shopper's searches, views, and purchases. Here, generation means predicting a next product, rather than writing a conversational answer. The model also receives timing information, so similar shopping histories can lead to different suggestions as circumstances and seasons change.
Shopify reports a 0.94% relative increase in Shop orders in an online A/B test of an August model version. Its engineering account connects the deployed system to a measured shopping outcome, while describing several training and infrastructure changes that contributed to the system as a whole.
