RPU to create a new era of media data value

The Challenge

udn needed a future-ready approach to strengthen both reader experience and advertising performance, while fully leveraging its vast content and behavioral data. Key challenges included:

• Limited Content Intelligence: Traditional keyword tagging and topic clustering were insufficient for identifying deeper semantic meaning, limiting personalized recommendations.
• Audience Understanding Gap: Without a unified view of reader interest and ad intent, udn lacked precision in predicting which users were most likely to engage with specific ads.
• Advertiser Expectation Shift: Brands increasingly demanded high-value audience packages and outcome-driven targeting, which the existing system could not reliably deliver.
• Data Freshness Constraints: Long update cycles prevented timely optimization, reducing the platform’s ability to react to rapid news cycles and advertiser needs.

To remain competitive, udn needed an AI-powered solution capable of transforming massive content assets and anonymous user journeys into actionable, monetizable data signals.

The Solution

udn partnered with Merkle to develop RPU —a comprehensive AI-driven media intelligence engine designed to revolutionize how content, readers, and advertising interact. Leveraging advanced deep learning, semantic analysis, and cloud engineering, RPU delivered a fully modernized data value system for udn.

1. AI Semantic Intelligence: Understanding Content at Scale

Merkle built a deep-learning–based Keyword & Entity Extraction Model using NER, POS, and TextRank to automatically interpret udn’s 600,000+ articles.
This enabled consistent recognition of:

• High-value entities (brands, products, people, locations, events)
• Ad-relevant topics and contextual signals
• Content areas with stronger commercial value

This formed the foundation for precise content classification and downstream modeling.

2. Interest Labeling Engine: Connecting Content to Advertising Intent

Using SBERT, Merkle projected 53 content topics and 25 ad categories into a shared vector space, allowing udn to measure true semantic relatedness between what users read and the ads they were likely to click.
Benefits included:

• Predictable ad-interest mapping
• Higher-quality audience segments
• Improved relevancy for both content and advertising

3. Click Prediction Engine: Forecasting User Behavior

To capture behavior patterns, Merkle deployed a DeepFM click prediction model, combining reading history, topic affinity, and ad interactions. The model assigned high click-probability labels to users, enabling udn to create powerful, performance-driven audience packages.

4. Real-World Media Experiments & Validation

udn executed two waves of advertiser testing across four placement types, validating RPU's commercial impact in live environments.
RPU-powered audiences outperformed traditional targeting in all test campaigns, demonstrating superior accuracy, engagement, and scalability.

Through RPU, udn transformed its extensive content and behavioral data into a predictive, monetizable, AI-first media engine, establishing a future-ready foundation for sustained media innovation.

The Results

Merkle’s RPU solution delivered transformative improvements for udn, surpassing all media performance benchmarks and validating AI-driven content and audience intelligence as a new engine for media monetization.

Topline Performance Results

• CTR (UU) +417%：Significantly increased unique-user engagement across all validated ad placements.
• CTR (Clicks) +200%：Demonstrated strong improvements in content–ad match quality and user interest relevance.
• 100% Win Rate：All five advertiser test campaigns outperformed existing targeting benchmarks.
• Mapping Rate +10%：Improved classification accuracy through enhanced semantic tagging and interest modeling.

Beyond performance lift, RPU strengthened udn’s long-term data infrastructure and operational efficiency:

• Reduced Data Latency：Content and behavioral data freshness improved by 2 days, enabling more responsive ad and content recommendations.
• Scalable Audience Products：25 high-value interest segments created for commercial use, enhancing advertiser targeting precision.
• Automated Content Intelligence：Semantic extraction and classification applied across 600,000+ articles, drastically reducing manual workload.

udn’s success was driven by Merkle’s ability to unite content understanding, user journeys, and ad intent into a cohesive AI-powered prediction system. By bridging semantic relevance with behavioral modeling, RPU allowed udn to deliver more personalized reading experiences, higher-quality traffic to advertisers, and a scalable framework for continued media innovation.