Case
AWS Personalize Optimisations for PBS
Who are PBS?
PBS is a US non-profit organisation that receives public funding and serves as a major provider of educational programming to public television stations in the United States. PBS’s sources of funding primarily come from member station dues and charitable donations from private foundations and individuals. Unlike other television networks, PBS does not rely on subscriptions or advertising to sustain its operations.
The Challenge
PBS, a widely recognised public broadcaster in the United States, is known for its diverse and engaging content. To make this content more accessible and relevant to its vast audience, PBS implemented personalised recommendations.
PBS aimed to leverage advancements in machine learning and AI to further enhance their consumer recommendations. Their vision was to make their personalised recommendations more dynamic, interactive, and reflective of the changing preferences of their audience. At the same time, PBS also sought to optimise their resources and improve cost efficiency.
PBS approached Merapar with a clear objective: to refine their recommendation system and maximise its performance while also managing costs effectively. The goal was to incorporate features like comprehensive A/B testing, more effective use of user interaction data, and exploring new functionalities of advanced recommendation engines. By achieving these objectives, PBS aimed to provide their viewers with an even more tailored and enriching experience while also improving operational efficiency.
The Solution
Our team, made up of seasoned data science engineers and Amazon Personalize experts, was brought on board to optimise the recommendation system. We embarked on this project by conducting an extensive analysis of the existing data set. Our goal was to understand the existing operations in depth and strategise improvements that would align with PBS’s objectives.
Implementing an agile methodology and the Scrum framework, we worked closely with the PBS team in a highly collaborative environment. Over five iterative sprints of two weeks, we made consistent progress towards our shared goals, regularly incorporating feedback to ensure that all advancements were rooted in effective and constructive communication.
With a deep understanding of the Video-On-Demand (VOD) domain, we identified key opportunities for optimisation. This insight led us to simplify the data pipelines and transition to using an Amazon Personalize VOD model. This not only improved precision metrics but also streamlined the recommendation system, enhancing its efficiency.
Recognising the importance of real-time user interaction data, we proposed necessary adjustments to client applications to incorporate an interaction event stream. This key enhancement improved the model’s freshness and overall performance of the recommendation system. With the integration of real-time user interactions, we were able to achieve a higher level of personalization, leading to a more enriched user experience.
The Results
The improvements we implemented in PBS’s recommendation system brought significant advancements in both efficiency and accuracy. By transitioning to an Amazon Personalize (media) framework, we not only streamlined the recommendation process but also ensured that PBS could easily integrate future enhancements from AWS, thereby setting the stage for continued advancement.
The enhanced pipelines contributed to cost efficiency, optimising resources and making the process more agile. An important development was the opportunity for a new context within the recommendation model, recognising the unique dynamics of TV, mobile, and web viewing. This greatly improved the personalization of content recommendations.
Incorporating user history into the recommendation process allowed us to offer more precise, user-specific content suggestions, thus improving viewer satisfaction and engagement.
Since going live with the new Recommendation Engine, PBS has seen a 10-18% increase in weekly click traffic across all platforms (web, mobile and OTT apps) on the presented shows over the manually curated collection. And they continue to see growth in engagement week-over-week..
In summary, this project provided PBS with a more efficient and personalised recommendation system that’s poised for future enhancements. The experience underscored the value of continuous innovation, and highlighted the potential of personalised recommendation systems in enhancing user engagement in the broadcasting industry.
Feedback from PBS
Molly Claverie: Senior Product Manager, PBS
“AWS Personalize as a service itself is already proving its value for us with the recommended shows collection on our website, surfacing shows that are not usually highlighted for viewers on our homepage and ensuring more relevant content for our viewers is front and centre. The click traffic increase over our manually curated collection on our homepage is a great indication that our viewers are finding what they want and exploring content generated from the Engine. We’re going to be rolling out our personalized collections on our App home screens soon and knowing how much traffic our viewers generate on their favourite shows and continue watching collections, I think personalized recommendations will be a huge hit on those platforms!”
Merapar was an invaluable team for us, and we would not have been able to launch in Q1 of this year without their help! We could easily have kept them busy for months to come, but they were critical to helping us better understand our data gaps, set up our new VOD Domain infrastructure, and adjust our Personalize configurations to meet our business goals. Not only were they incredible experts of the service, but they were friendly, diligent, and flexible partners throughout our time together. They graciously left us with amazing documentation and a list of further suggested improvements that we will use to influence our roadmap for months to come.”
Feedback from Merapar
Simon McGrath: VP Strategy & Portfolio
“PBS approached us with a very specific challenge to help them fully launch their personalized recommendation service with a refined data model and workflow. We brought in a highly experienced team, and over a short and intense period of five sprints, including on-site interaction with the PBS team, we implemented a production platform that is already delivering great results.”
