PLATFORM
TV, iOS/Android
WHAT IS AI TOPICS
It’s a Gen AI based discovery assistant that provides Prime Video customers a list of personalized topics and recommends them specific content to stream for a selected topic.
TIMELINE
Aug 2024 - Nov 2024
ROLE
I was the sole designer on this fast-paced project and conceptualized the experience from 0 to 1. I reported to my skip manager for design feedback and collaborated with the design system team on branding.
BUSINESS OPPORTUNITY
Using AI to transform the future of streaming
With AI technology taking over diverse industries and simplifying menial tasks, there’s a huge opportunity to leverage it when it comes to the streaming and entertainment world. Prime Video sees a huge opportunity to help its customers discover new content from its ever-expanding catalog and drive more future studio partnerships. Today, only 43% of Prime Video’s discovery sessions end with a stream. Given the size of Prime Video’s ever-expanding catalog, there should be something new for everyone to watch.
The business goal here to improve customer engagement and eventually hours streamed by making it easier for customers to find something they want to watch. Furthermore, the initial launch will help us understand if Prime Video customers find a feature like this useful and leverage the learnings to expand the Assistant's capabilities over time.
Prime Video’s vast content offering, spanning multiple partnerships
CUSTOMER PROBLEM
Majority of Prime Video visitors fail to find something to watch and end up leaving the platform.
Only 43% of Prime Video’s discovery sessions (ones that don’t end with a re-engagement stream) end with a stream. This indicates that many customers try to find something new to stream but don’t. Given the size of the Prime Video catalog (TV titles), there should be something new for everyone to watch. Customers agree, with “Finding content matching specified criteria” rated as ‘not good enough’ by 15% of customers and PV ranking the worst out of four competitors (Apple TV+, Hulu, MAX, and Sky TV) when survey respondents are asked, “How easy is it to locate videos that match your specified preferences?”. Additionally, ~23% of customers drive ~70% of hours streamed on PV, and it is critical we are able to move less frequent streamers up the engagement curve and turn them into power streamers.
Oftentimes, customers only have a general idea of what they want to watch based on different factors like a mood (e.g., “I want to watch something uplifting”), a desire to learn (e.g., “I want to watch a documentary“), or an interest (e.g., ”I want to watch a time-bending sci-fi flick“). This general idea can change every time a customer turns on their TV. Unfortunately, today customers either have to scroll through a sea of content hoping that a suggested movie or TV show that matches their preferences or they need to know what exact to type if they decide to go looking for it in Search. This results in customer frustration as they spending a majority of their session time in figuring out what to watch or abandoning the platform unable to find something.
DESIGN SOLUTION
An interactive, personalized discovery assistant to help Prime Video customers find their next watch
Prime Video customers will be served with a list of topics through this feature, where the topics are generated using GenAI. These topics will be generated for every user parsing PV’s vast catalog and eventually help customers discover new content. The customers will see a mix of topics including - trending as per their area, seasonal and personalized ones. Furthermore, these topics will be different for every session and customer can choose to refresh them.
COMPETITIVE ANALYSIS
Understanding industry patterns
I began by studying current industry trends for GenAI products to understand how users' mental models are shaped by existing technologies. This research informed several of my design decisions. Notably, this is the first time we're incorporating GenAI into a TV experience—while GenAI features are well-established on mobile and web platforms, applying them to TV comes with unique challenges. It's crucial to adapt these features to fit the specific nuances and limitations of the TV interface.
IDEATION PHASE
Design iterations
PRIME VIDEO’s HOMEPAGE MODULE
For the homepage module, I focused on defining the taxonomy, visual representation, and density of AI cards. To ensure the design resonated with users, I consulted with a PV researcher to understand how customers currently engage with tiles on the app. Key insights included that users respond more to images than text, often ignore carousel titles, and have moods that influence what they want to watch.
My initial design presented two options: Option A, a wall of text that exposed more topics at a glance but felt robotic, and Option B, a more visually engaging approach with fewer topics visible at once. I worked with a Data Scientist to define what constitutes a customer’s "interest" based on their watch history, search behavior, local trends, and seasonal content. This helped guide the taxonomy of the AI cards.
Although the product team initially favored Option A for its density, I convinced them that a more visually engaging approach would be more effective in drawing user interest. Additionally, I used purple and blue to differentiate the AI assistant, initially named Beacon, giving it a distinct branded identity.
RESULTS PAGE FOR SELECTED AI TOPICS
The debate between using a grid or carousel layout was a key discussion, with the product team advocating for a carousel UI that allowed unlimited exploration and the ability to show trailers. My perspective was to offer a more focused, exploratory experience through a grid layout, enabling users to quickly scan results and easily pivot if they didn't find anything of interest.
To inform this decision, I collaborated with a designer from the search team to analyze user behavior in Prime Video search, aiming to align the design with existing user mental models for navigating results. Drawing on my deep understanding of Prime Video’s product ecosystem, I advocated for reusing existing frameworks and mental models rather than reinventing solutions. This was a challenging push with the product team, but I also took the lead in defining the overall page structure, including the number of result rows and when pivot points should appear. By leveraging the existing pattern across Search, the engineering team was also able to accelerate the timeline for implementation. Additionally, I considered the effectiveness of video results by showing watched titles but ranking them lower, and addressed how to handle entitled versus unentitled results.
AI TOPIC’s IMMERSIVE LANDING PAGE EXPERIENCE
Based on research, we know that customers engage more with a movie or TV show's title artwork, as it's a key visual cue that sparks interest. For the AI Topic’s landing page experience, I explored multiple layout options. The initial iterations included Option A from Iteration 1 that is high density and quick visibility vs a second mosaic view for a more visual approach while sacrificing breadth for simplicity. Similar to that challenges identified during the homepage component, I pushed the product team to consider a more visual option to prevent information overload. The intention is also to have a UI that’s consistent with the homepage module.
There were spicy debates as I tested around 8 concepts for the landing page, experimenting with different densities of topics, text versus image-based cards, and poster usage. To push the product team, I created a prototype showing that overly dense, text-heavy cards could hurt usability, leading to more clicks and user frustration. I worked closely with the product and tech teams to align on incorporating some image cards to add visual interest and break up the text monotony. Ultimately, we settled on the bottom-right variation, which balanced higher-density text cards with some image-based ones for a more engaging and usable experience.
DESIGN DELIVER PHASE
Visual Languae and Branding
I was intentional about using blue and purple in the feature, associating these colors with trust and authority, while the gradient symbolized change. Initially, I considered giving the Gen AI agent a distinct personality, but after design leadership reviews and aligning with product on how the product could evolve in the future, I ultimately chose a more generic approach to allow for flexibility. I collaborated closely with a designer from Prime Video's design system and a marketing designer to define the overall color scheme, providing guidance to align with both Amazon's broader branding and PV's identity. My guidance was to choose something general like sparkles with the rationale that it should represent magic, a common industry symbol for AI products.
Additionally, to prevent collusion and maintain uniqueness, I conducted a quick audit of AI product colors within Amazon, identifying commonalities while ensuring Prime Video’s AI discovery assistant stood out.
FUTURE SCOPE
Proposed Mobile Experience
I designed the mobile experience to align closely with the TV version but incorporated additional features, such as voice-enabled input and keyboard support. While the TV experience could potentially integrate Alexa in the future, the mobile app includes a floating entry button for easy access, ensuring it doesn't interfere with other elements like main navigation. The design also emphasizes voice and prompt input, with condensed result cards optimized to display quick actions and a streamlined view of video results.
IMPACT
AI Topics preferred over scrolling on homepage for discovering new titles
This feature was recently launched to 20% of Prime Video’s US customers with positive feedback. Around 200 customers gave AI Topics a 4.04 / 5 customer satisfaction score, and AI Topics has increased customer’s satisfaction for discovering content on PV by +0.29 (from 3.92 / 5 to 4.21 / 5). Additionally, 61% of customers agreed that they preferred using AI Topics over scrolling on the homepage, and 65% said they could see themselves engaging with AI Topics repeatedly as a means of finding new titles to watch.