#AI

From Clicks to Conversions: Architecting Shopping Conversion Candidate Generation at Pinterest

From Clicks to Conversions: Architecting Shopping Conversion Candidate Generation at Pinterest
01

Summary

From Clicks to Profits: How Pinterest Redefined Shopping Conversion Ads with AI

Innovating Ad Performance and User Experience through Parallel DCN v2 and Multi-Task Learning

This article details Pinterest's journey in architecting a specialized candidate generation model for shopping conversions. It explores how they evolved from engagement-based retrieval to a sophisticated system capable of capturing high-intent purchase signals among 600 million users.

  • 01Utilized Parallel DCN v2 and MLP to eliminate information bottlenecks and capture richer feature interactions.
  • 02Implemented a log-based re-weighting function for click duration to filter noise from engagement signals.
  • 03Introduced an advertiser-level loss function to stabilize training against high variance at the individual Pin level.
  • 04Evolved from a multi-head to a unified multi-task architecture to directly benefit serving performance.

RECOMMENDATION

Highly recommended for ML engineers focusing on conversion optimization in large-scale recommendation systems and architects seeking to improve retrieval quality.

The Problem

Pinterest faced challenges in optimizing for lower-funnel conversions because offsite conversion signals are sparse, noisy, and delayed compared to onsite engagement signals.

The Solution

The team implemented a two-tower model with parallel DCN v2 and MLP cross-layers, utilizing weighted engagement data based on click duration and an advertiser-level loss function.

The Result

The system achieved a 2.3% increase in shopping conversion volume, a 3.1% improvement in RoAS, and an average +42% increase in recall@100 for conversion tasks.

Trade-off

Addressing data sparsity required more complex multi-task balancing and advertiser-level granularity, which increased architectural complexity compared to engagement-only models.

03

Key Concepts

Concept · 01

DCN v2 (Deep & Cross Network v2)

A model architecture that combines a cross network for explicit feature interactions and a deep network for implicit patterns.

  • Applied a parallel design to allow both networks to learn directly from input features simultaneously.
  • Enabled the construction of higher-order feature crosses without losing information from preceding transformations.
Concept · 02

Two-Tower Model

A retrieval architecture where user and item features are encoded separately into embeddings for efficient similarity matching.

  • Optimized for high-scale retrieval by separately processing User and Pin towers.
  • Used as the foundation for the shopping conversion candidate generation system.
Concept · 03

Multi-Task Learning (MTL)

A learning paradigm that trains a model on multiple related tasks simultaneously to improve generalization and handle data sparsity.

  • Combined conversion signals with engagement data to stabilize training for sparse labels.
  • Adopted a unified single-head architecture to better align with the natural granularity of conversion signals.