Highlights

Condé Nast was able to quickly deploy a multivariate testing initiative for improving customer experience, achieving low single-digit millisecond response and significantly reduced reprocessing times with a solution powered by Cassandra.

50% faster delivering digital content

30% improvement in click-through

<4ms response times

20 brands, 50M daily visits, 100M unique visitors per month.

Condé Nast is a premier media company renowned for producing high-quality content for the world’s most influential audiences. Attracting more than 144 million consumers across its industry-leading print, digital, and video brands, the company’s portfolio includes some of the most iconic titles in media, including Vogue, Vanity Fair, Glamour, GQ, The New Yorker, and Wired.

Challenge

Condé Nast Inc. is an American mass media company founded in 1909 by publisher Condé Montrose Nast. Today, the company is owned by Advance Publications. In 2016, Condé Nast announced the launch of Condé Nast Spire, a new division that focuses on finding links between consumers’ purchasing activity and content consumption via Condé’s own first-party behavioral data.

Today, Condé Nast’s business objective is to increase subscription rates by improving the customer experience and customer engagement model. In order to determine how to better engage its customers, Condé Nast launched a multivariate testing initiative with the goal of fully understanding its user base and the types of content, web layouts, and visual displays that appeal to each target segment.

In addition, Condé Nast wanted to leverage the data gathered in its multivariate testing initiative to provide personalized content and recommendations to web visitors. This, they hoped, would translate to a more engaging experience and increased online subscribers. Inspired by Uber’s AI and machine-learning project, Michelangelo, Condé Nast set out to build its own machine-learning workflow with a Feature Store backed by Cassandra.

Solution

Condé Nast needed a powerful database solution and narrowed its search to two NoSQL databases. Cassandra won during a benchmark comparison where the team tested each solution’s ability to scale to match Condé Nast’s traffic from more than 20 brands, 50 million daily visits, and more than 100 million monthly unique visitors.

Results

Response times proved to be less than four milliseconds for 6,000 requests per minute, allowing Condé Nast to execute new website tests fast and as-needed. Moreover, the Feature Store empowers data science teams to make successful real-time predictions, deliver targeted marketing content more effectively, and increase user engagement on their web and mobile properties.

Reprocessing time has improved by 650%, meaning that Condé Nast can double the number of models and projections stored. Feature Store read latency is now less than 4 milliseconds for 1,800 requests per minute, and write latency is less than 10 milliseconds for the same amount of requests.