Pinterest integrates real-time browsing context into sequential ad recommender models, boosting Related Pins candidate relevance by 275–300% and ROAS by 0.7%
The sequential candidate generator inferred user embeddings purely from offline historical behavior, leaving it blind to what a user was actively browsing at ad-serve time — a critical gap on contextual surfaces like Related Pins and Search.
How it works
Common implementation structure
How this type of workflow is generally built, generalized across documented cases — not tied to any one vendor's stack. Click any stage to read what happens there. Specific products that implement these stages appear in “Tools commonly seen” below.
Stage 1 · Historical sequence offline inference
The Transformer encoder processes the user's historical event sequence offline and its last hidden state is stored in the feature store.
Tools used
TransformerMulti-Layer Perceptron (MLP)feature store
Outcome
The Contextual Sequential Two Tower Model achieved a 3x to 10x increase in Recall@K, lifted median candidate relevance by ~275–300%, improved the Related Pins ads relevance metric by 1.08%, delivered 2x more ad candidates to impression, and drove ~0.7% ROAS lift overall (~1.4% for top countries).
What failed first
In the previous production system, fewer than 1% of impressions on Related Pins were attributed to the sequential candidate generator because candidates lacking real-time context could not survive downstream ranking and auction.