Pinterest uses a multi-layer AI system to predict which users are most likely to take a specific action after seeing an ad — such as making a purchase or adding something to a cart. This prediction shapes which ads get shown to which users and at what price. The system has been continuously upgraded over several years, shifting from simpler rule-based tools to deep neural networks.

The current system uses a Multi-Task Ensemble Deep Neural Network — a type of machine learning architecture that learns to predict multiple outcomes simultaneously (clicks, purchases, add-to-cart events) rather than optimizing for just one. It incorporates Deep & Cross Network v2 (DCNv2), a technique for capturing complex interactions between user and ad features, and a transformer-based component for modeling sequences of user actions over time. The model was described in a January 2024 engineering post, though the system has been iterating for several years and serves hundreds of millions of users. The outputs directly determine which ads are shown and what advertisers are charged.