Recirc Using Content Information

Project Overview

An internal machine learning powered recommendation widget designed to increase session length and engagement by serving personalized content suggestions to anonymous users. The system leveraged automated tagging, machine learning, and contextual signals to deliver high-value recommendations without relying on personal identifiable information.

Business Challenge

At the time this project was built, off-the-shelf recirculation tools were not yet available. The company needed a scalable solution to keep users on-site longer while maintaining privacy. Key challenges included:

• No existing tools for anonymous, session-based content personalization.
  
  

• Risk of user drop-off after consuming a single piece of content.
  
  

• Inability to leverage PII due to privacy requirements.
  
  

• Seasonal and regional factors influencing content relevance that weren’t captured by traditional “most popular” feeds.
  
  

My Role

As Senior Technical Product Manager, I:

• Defined requirements for a privacy-first, machine learning–driven recirculation system.
  
  

• Partnered with engineering to build a tagging and recommendation pipeline.
  
  

• Guided feature prioritization, including seasonality and geolocation signals.
  
  

• Oversaw testing and deployment across multiple brand sites.
  
  

Solution

We built a machine learning–driven recirculation widget that analyzed content consumption behavior during a single session and dynamically recommended the most relevant next articles. Core features included:

• • Content Tagging: Used a Google automated tag assistant to classify site content by main topics.
    
    

  • Session Behavior Tracking: Logged which tagged pages a user visited during their session.
    
    

  • ML Recommendation Engine: Compared visited pages against all tagged content to compute similarity scores and identify the best next content to surface.
    
    

  • Real-Time Display: Once enough user behavior was captured, the widget appeared on subsequent pages to provide tailored suggestions.
    
    

  • Seasonality Scoring: Boosted scores for content relevant to seasonal patterns (e.g., holiday recipes).
    
    

  • Geographic Relevance: Used IP-based general location to promote content aligned with local interests.
    
    

  • Privacy First: Fully anonymous — no PII required or stored.
    
    

Business Impact

• • Significantly improved session length and user retention by surfacing more relevant content.
    
    

  • Enabled personalized experiences without the need for personal data collection.
    
    

  • Increased visibility and distribution of seasonal and regionally relevant content.
    
    

  • Served as one of the company’s first applications of machine learning for real-time user engagement.
    
    

Technical Highlights

• Google automated tag assistant for large-scale content classification.
  
  

• Machine learning model for similarity scoring across tagged content.
  
  

• Real-time session tracking and recommendation pipeline.
  
  

• Seasonal and geolocation scoring integrated into final ranking algorithm.
  
  

• Privacy-first design — no reliance on user PII.