Warner Bros. Discovery (WBD) is a global leader in media and entertainment, offering a diverse content portfolio across television, film, and streaming. As user demand for personalized content continues to grow, the company decided to enhance its user experience by improving its artificial intelligence and machine learning (AI/ML) inference infrastructure. In this process, they adopted AWS Graviton processors and Amazon SageMaker AI instances, achieving significant cost savings and performance improvements.
With over 125 million users worldwide, WBD's personalized recommendation system needs to run efficiently in real-time environments. To address increasing user demand and cost pressures, the company began migrating to Graviton instances with the support of AWS. Through this approach, WBD not only reduced infrastructure costs but also improved inference speed, achieving an average saving of 60%. In some cases, particularly in catalog ranking models, costs were reduced by as much as 88%.
In terms of inference speed, WBD has achieved significant results. With the optimized infrastructure, P99 latency decreased by 7% to 60% across different models, with XGBoost models experiencing a reduction of up to 60%. This performance improvement enables users to enjoy faster and more accurate content recommendations, enhancing user engagement and retention.
The migration process for WBD went smoothly, taking only one month from initial testing to full deployment. The company plans to continue migrating more recommendation systems to Graviton instances to further improve operational efficiency and reduce costs.
Key Points:
🌟 Achieved 60% cost savings: WBD significantly reduced the operating costs of its personalized recommendation system by migrating to AWS Graviton instances.
⚡ Improved inference speed: P99 latency decreased by 7% to 60% across different models, providing users with faster content recommendations.
🔧 Smooth migration process: From initial testing to full deployment, it took only one month, ensuring efficient project execution.
