Building Holiday Finds: How Pinterest Engineers Reimagined Gift Discovery

Megan Blake, Usha Amrutha Nookala, Jeremy Browning, Sarah Tao, AJ Oxendine, Siddarth Malreddy

Overview & Context

The holiday shopping season presents a unique challenge: helping millions of Pinners discover and save perfect gifts across a vast sea of possibilities. While Pinterest has always been a destination for gift inspiration, our data showed that users were facing two key friction points: discovery overwhelm and fragmented wishlists. With 85% of weekly US Pinners having made a purchase based on Pins from brands¹, we saw an opportunity to create a more streamlined gift discovery experience that meets Pinners where they already are — at the early stages of shopping inspiration.

The Challenge

Holiday gift shopping traditionally involves juggling multiple browser tabs, wishlists, and recommendations. Pinterest users actively look for new brands and ideas, with 96% of top searches being unbranded², showing Pinners are open to discovering fresh products. The engineering challenge was clear: how could we create a unified experience that maintains the serendipity of Pinterest discovery while adding structure to the gift shopping journey?

Our solution needed to:

Leverage our existing personalization infrastructure while introducing gift-specific optimizations
Create a seamless saving experience that automatically generates shoppable wishlists
Build a UI that feels fresh and distinct from our standard feed experience

Technical Foundation

We approached this challenge by building on three core technical pillars:

Personalization Stack: Rather than building a new recommendation system from scratch, we extended our proven Homefeed architecture with gift-specific candidate generators and ranking signals.
Dynamic UI Framework: We developed a new Structured Feed Framework that allows rapid iteration on UI components while maintaining platform-specific optimizations for iOS, Android, and Web.
Unified Logging System: We implemented comprehensive engagement tracking that helps us understand how users interact with gift content differently from standard Pins.

Personalization Stack

Building a Gift-Optimized Recommendation System

The success of Holiday Finds hinges on our ability to surface the right gift ideas at the right time. Rather than building an entirely new recommendation system, we strategically adapted our proven Homefeed architecture through a three-stage pipeline: candidate generation, ranking, and blending.

Candidate Generation: Smart Sourcing of Gift Ideas

Our first challenge was identifying “gift-worthy” content from Pinterest’s vast Pin corpus. We developed two specialized candidate generators:

Proven-Shoppable Generator: This system identifies Pins with high purchase intent signals, focusing on items that historically perform well as gifts. We analyze metrics like save-to-purchase conversion and gift board additions to surface promising candidates.
Inspirational Corpus Generator: Launched post-initial release, this generator enriches recommendations with contextual gift ideas. It employs semantic filtering with a “seasonality” signal, ensuring that inspiration aligns with current gift-giving occasions.

Ranking: Personalized Gift Relevance

The ranking stage employs our Homefeed Pinnability model with gift-specific optimizations:

Real-time user sequence features capture immediate shopping intent
Integration with our inclusive AI systems ensures diverse gift recommendations across various styles and preferences

Blending: Crafting the Perfect Mix

The blending system orchestrates the final feed composition through three key mechanisms:

Interest-Based Diversification: Our Determinantal Point Process (DPP) algorithm prevents recommendation tunneling by applying interest-penalties to similar items.
Gift-Specific Filtering: A post-ranking filter removes utilitarian products while elevating items with strong gift signals.
Module Integration: Specialized shopping modules are interwoven throughout the feed, providing structured entry points for different gift categories.

Figure 1: Holiday Finds Backend Design

Autogenerated Wishlist Boards

Smart Wishlist Generation

A key innovation of Holiday Finds was its ability to automatically create and populate shopping-focused boards. Rather than requiring users to manually create and organize boards, we developed a system that reduces friction in the gift discovery journey.

Technical Implementation

The wishlist generation system operates through several coordinated components:

Monitors first save action from Gift Guide surfaces
Validates user eligibility (no existing wishlist for current campaign)
Handles edge cases like archived or soft-deleted boards

2. Board Creation Pipeline

Asynchronously creates new board with “Wishlist” designation
Maintains internal flags to track wishlist status regardless of user renaming
Implements soft deletion and archive states that pause automatic saving

3. Quick Save Integration

Introduces Pin Icon Grid Save (PIGS) for one-tap saving
Maintains save state persistence within Gift Guide sessions
Handles multi-board scenarios and undo functionality

Performance Impact

The system has demonstrated promising results aligned with Pinterest’s shopping-oriented audience:

Generated significant wishlist adoption, helping Pinners organize their gift ideas
Streamlined the saving experience for holiday shoppers
Drove higher engagement, reflecting Pinterest’s strength as a platform where 80% of weekly Pinners say they feel inspired by the shopping experience³

Edge Case Handling

We implemented robust handling for several complex scenarios:

Board deletion/restoration flows
Collaborative board permissions
Name collision resolution
Archive state management
Cross-platform state synchronization

This infrastructure provides a foundation for future automatic organization features beyond holiday shopping, demonstrating how we can reduce friction in content organization while maintaining user control and expectations.

UI Implementation

Crafting a Distinctive Shopping Experience

The Holiday Finds interface needed to feel special while maintaining Pinterest’s familiar comfort. We developed several key innovations to achieve this balance:

Dynamic Hero Header

We built a new dynamic header system that creates an immersive entry point to gift discovery:

Adaptive Imagery: The header automatically selects and blurs high-performing Pins from the user’s feed, creating a cohesive visual story that updates with each refresh.
Platform-Specific Optimization: We implemented distinct module structures for mobile and web, with iOS/Android utilizing a dual-module system while web maintains a unified approach.
Smooth Transitions: We engineered careful handling of status bar interactions and scroll behaviors to ensure the experience feels native on each platform.

Structured Feed Framework

To support rapid iteration and maintain consistency across platforms, we developed a new Structured Feed Framework. Key features include:

Configurable Module Headers: Easy definition and modification of module properties without code changes
Cross-Platform Compatibility: Automatic handling of platform-specific UI requirements
Performance Optimization: Smart image loading and caching strategies to maintain smooth scrolling

The framework’s flexibility proved invaluable during development, allowing us to:

Quickly test different header configurations
Adjust module layouts based on user feedback
Deploy platform-specific optimizations without duplicating code

Holiday Finds on Web

Figure 2: Holiday Finds Recommendation Landing Page

Tab Ranking System

Solving the Cold Start Problem with Dynamic Tab Ranking

One of our key challenges was introducing a new shopping surface while maintaining the integrity of Pinterest’s navigation system. We developed a two-phase tab ranking strategy that balances discovery with personalization.

Bootstrap Phase

To ensure users could discover Holiday Finds, we implemented a fixed-position strategy:

Three-day bootstrap period with Holiday Finds locked to position 1 (immediately after “All”)
Existing Board More Ideas tabs maintain their engagement-based ranking
User behavior tracking begins immediately to inform future positioning

This approach gave us two key advantages:

Guaranteed visibility during the critical discovery period
Clean data collection for long-term ranking optimization

Engagement-Based Dynamic Ranking

After the bootstrap period, we transition to a dynamic ranking system that considers multiple engagement signals:

- Pin saves to boards

- Direct Pin clicks

- Social interactions (likes, comments)

- Feed engagement (10+ impressions)

The ranking system updates daily through a batch process, helping us balance computational efficiency with responsiveness. To avoid latency issues, we built a smart caching system that only updates tab positions when the bootstrap period has conclusively ended.

Logging & Analytics

Building a Data Foundation for Shopping Innovation

To measure success and enable continuous improvement, we implemented a comprehensive logging system that extends our existing Homefeed analytics infrastructure.

Unified Event Tracking

We designed our logging system to capture the unique aspects of gift discovery behavior:

Feed-level metrics: unique feed IDs track content distribution and engagement patterns
Module-specific interactions: granular tracking of how users engage with different shopping modules
Tab transition analysis: understanding how users move between Holiday Finds and other Pinterest surfaces

Implementation Strategy

Rather than building a separate logging system, we extended Homefeed’s Local Navigation logging capabilities:

Reused existing tab action tracking infrastructure
Added gift-specific event types and attributes
Implemented validation queries to ensure data quality

This approach allowed us to:

Launch quickly with proven infrastructure
Maintain consistency with existing analytics
Enable seamless cross-surface analysis

Future Work

Impact and Looking Ahead

Results

The Holiday Finds launch has demonstrated strong initial success:

Improved engagement with wishlists compared to standard boards
Strong adoption among Gen Z users, our fastest growing audience³ that makes up 42% of our global user base³
Successful integration of curated gift guides globally, helping Pinners connect with brands at their first moment of inspiration

Technical Learnings

Several key insights emerged from this project:

The Structured Feed Framework proved highly adaptable, supporting rapid iteration across platforms
Our two-phase tab ranking strategy effectively solved the cold start problem while maintaining personalization
The reuse of Homefeed infrastructure significantly accelerated development while maintaining reliability

Future Directions

We’re excited to build on this foundation in several ways:

Exploring different ranking models for specific tab types
Investigating real-time ranking updates for increased responsiveness
Testing new engagement signals for better personalization

2. Shopping Experience Enhancement

Expanding the dynamic header system to other Pinterest surfaces
Developing new shopping-specific modules
Further optimizing the gift discovery algorithm

3. Platform Innovation

Extending the Structured Feed Framework to support more complex interactions
Building new tools for rapid shopping surface development
Improving cross-surface content discovery

Acknowledgement

Kurby Gebremedhin, Samantha Lee, Jesse Andersen, Kirsten Browne, Jiaqi Tong, Ianyu Feng, Tian Li, Josh Arriola, Elizabeth Cheng, Jazz Ernest, Emma Li, Zihao Chen, J.J Hu, Raymond Hsu