Condé Nast Delivers Relevant Content 90% Faster with MongoDB Atlas

THEIR CHALLENGE

Leading readers to more relevant content

Global media company Condé Nast wanted to improve the user experience for the millions of readers who engage with its brands, from GQ to Vogue. Specifically, the company sought to optimize its “read more” functionality, which guides visitors to similar content across its more than 70 websites. However, Condé Nast’s vast repository of text, audio, video, and images made this project a highly complex endeavor.

To manage this complexity, Condé Nast’s engineering team built three separate recommendation pipelines on Elasticsearch. However, it wasn’t feasible to scale unique pipelines for each website and asset class.

The need for change came to a head when Condé Nast introduced a new open-source embedding model to Elasticsearch to increase recommendation quality. The results were disappointing: response times shot up, cost and complexity spiked, and the embedding relevance dropped. “We needed a solution that was going to scale much better,” said Chris Chen, Global Vice President of Architecture and Governance at Condé Nast. “We already had storage on MongoDB, and it made sense to simply use MongoDB to process vectors instead of sending the data to an external pipeline.”

OUR SOLUTION

Using MongoDB to maximize recommendation accuracy for Condé Nast

Condé Nast moved all content from across its portfolio to MongoDB Atlas, a fully managed, multi-cloud database service. From there, it ran its content through Voyage AI embedding models, which offer state-of-the-art retrieval accuracy, to generate image vectors. The resulting embeddings—approximately 2 million vectors, with a future goal to vectorize all content—were then stored in MongoDB Atlas Vector Search to generate recommendations through image search. For Condé Nast, Vector Search handles 1,500 queries per second for recommendations and 500 queries per second for search.

From the very beginning of its relationship with MongoDB, the Condé Nast team felt assured in its decision to expand its stack. “After reaching out, a MongoDB expert was with us within 48 hours,” said Tina Varughese, Director of Engineering for Customer Data Products at Condé Nast. “The proof of concept took only a couple of weeks.”

Continual, responsive collaboration was another highlight of the implementation process. “When we have an issue or need some additional capabilities, the MongoDB team has always been very supportive in finding solutions,” said Chen. “Their architects and engineers have done a great job augmenting our staff and helping us achieve our objectives.”

OUTCOME

Delivering more relevant content to millions faster

With the optimized recommendation engine, Condé Nast delivers more relevant content to readers faster. Per early measurements, the MongoDB solution reduced latency by 90% and cut operational costs by 65%. Click-through rates across Condé Nast’s websites rose to 35%, indicating more accurate recommendations. Further, the solution freed Condé Nast’s engineers to innovate without being interrupted by system outages or latency issues.

By migrating to MongoDB, Condé Nast set the stage for future innovations, such as hybrid search capabilities and other AI use cases. It is also testing Voyage AI’s capabilities to improve search across all content types in addition to images. “This project with MongoDB was one of the most impactful projects that I’ve done at Condé Nast,” said Varughese. “No matter what future proof of concept we want to do, now we have a stable system to experiment on.”