Content Discovery & AI Part 1: Video streaming
Media and entertainment has evolved over the last few decades. With the arrival of computers, digitization of content, and the rise of smartphones and social media, we have witnessed the largest content explosion. Content has not just grown in volume but also in shape, size, and format. We can listen to podcasts and audiobooks on apps like Spotify, watch short form videos on aps like Tiktok, view movies and tv shows on streaming platforms like Netflix, and more. With Generative AI making content generation even easier, content will grow even faster, and discovery will continue to be a major challenge. In this series on content discovery, I will explore ideas to improve discovery for users on video streaming platforms, audio platforms, and short form video apps. I will also talk about advancements in AI and how it’s impacting content generation, editing and publishing, and discovery.
Video streaming
In this article (Part 1 of the Content Discovery series), I will focus on video streaming platforms like Netflix. I will first talk about viewers and their top problems, and then share a few ideas to solve these problems I will only focus on platforms like Netflix, HBO and Apple TV, and exclude cable television. Cable television is declining in viewership and also follows a very different format. Below, I will begin discussing the key use cases for viewers.
Viewer needs and use cases
(1) I know exactly what I want to watch: This is a use case for viewers who are fans of a movie or tv show or a live sports event. They probably want to watch it as soon as it is released. The main problems for these viewers are: (a) finding content across the various services inside the streaming platforms (especially marketplaces like Prime Video and Apple TV) can be tough and search often does not provide a seamless experience; (b) users often do not get notified on time that their favorite show released a new episode or their favorite director or actor’s movie is available on their streaming service; and (c) they are not able to watch offline or in certain geographies or languages.
(2) I have some time to kill and I want to watch something: These are viewers who want to watch something now but not invest too much time in finding it. It could be before bedtime, or on Sunday afternoon, or on a plane or cab ride. They know what they like and they want something that sort of falls within that realm. It could even be something they have already seen and loved it. Key problems for these viewers are: (a) finding content takes more time than they have and the process can be exhausting; (b) there isn’t sufficient information or signals to make decisions; © the content is not available in a specific language, device or geography
(3) I want to discover something new and interesting.These are viewers who know what they like but they want to try something new and different. They want to explore and be surprised and delighted. They are open to investing time in exploring. Key problems for these viewers are: (a) many streaming platforms are not designed for enabling discovery; (b) often, the homepage is cluttered and it is overwhelming to sift through all the recommendations; © viewers do not feel in control of the discovery process. There is immense value in enabling exploration and discovery on content platforms (as also shown in this Google research study here).
AI advancements
Artificial Intelligence spans a wide array of technologies including rules based systems, machine learning (ML), deep learning, generative AI (GenAI) and more. Traditionally, technology companies have used rule based systems and ML (and in some cases, deep learning) to power search and personalization on software apps and systems. With recent advancements in Large Language Models (LLMs), Generative AI has come to the forefront with the technology to generate text, image, audio and video. GenAI companies have developed chatbots that can engage in human-like conversations, help users accomplish tasks and improve their productivity. Many users who used Google search in the past are now using ChatGPT, Perplexity and Gemini for the same tasks, which are better and faster at many tasks and queries.
GenAI will change how search engines work as users will soon expect all search systems to understand and respond to long and complex queries, and have human like conversations. Google is already making changes to incorporate AI’s in it’s search as described here. GenAI is also expected to improve recommendation systems by better matching user needs with the right content and enabling creative and personalized presentations of content. For example, Amazon here talks about how they now show personalized titles on products based on what the user search query and their purchase history and preferences.
Designing solutions
Based on our analysis above, key user problems in video streaming content discovery are: (1) too much content shown which is overwhelming and hard to sift through; (2) lack of quality markets and good previews; (3) lack of a human like chatbot experience; (4) no live feedback ingestion based on user engagement; and (5) sub-optimal use of human expertise. Below, I will discuss solutions to address these.
Pillar 1: Simplify the interface and reduce clutter
Streaming platforms should simplify their interface to make it less cluttered and overwhelming. Instead of showing tens of carousels with individual movies, shows or live sports, platforms should reduce and focus on showing less recommendations that are more personalized for every user. They should experiment with larger imagery and on-hover videos to make it easier for viewers to understand and engage with content. Netflix has simplified their interface along these themes in the last few years but it is still quite cluttered. However, others have not invested in this area.
Pillar 2: Add markers and differentiators to help users make faster decisions
Scrolling through movies and shows to find the one you love is not easy on streaming platforms and puts a lot of burden on the user. I recommend we experiment in the following three areas to improve decision making. One, add ratings and reviews from trusted reviewers like Rotten Tomatoes or Imdb, and experiment with adding in-platform user video and text comments. Two, add viewership numbers (how many viewers have watched this) and qualitative signals such as Most watched, In Top 100. Third, leverage existing movie trailers or use GenAI to experiment with personalized movie trailers.
Pillar 3: Leverage AI to help users talk to a movie, shows and sports expert
With GenAI chatbots making the waves, more users are flocking to ChatGPT, Perplexity and the likes to do research, generate ideas, plan travel, find products and learn new domains. These chatbots are available on text and voice, and users can upload documents, images and videos when needed. Imagine having a conversation with a streaming-bot which has adeep knowledge of movies, tv shows, live sports and other video content. Viewers can simply say or type what they are looking for. For example, “i want something light and easy to watch for an hour, something like friends ot 30 rock”, or “i am in the mood for a horror or scifi thriller movie like wicked, something from the last few years”, and you can get a few tailored recommendations based on that. You can interact with the chatbot and ask follow up questions. For example, “i like these but can you give more like the first movie you recommended”, or “do you any movie from this genre with Tom Cruise in it”. This can simplify discovery and make the process more human and delightful.
Pillar 4: Improve personalization to incorporate user feedback real-time
One complaint users often have across content platforms is that their signals like views and likes are not accounted for in their recommendations. Also, recommendations for new users are generic and not updated real-time or often even after multiple sessions causing frustration. I recommend investing in deep customer understanding to capture user signals and leveraging these signals in recommendation models. For example, understanding why a user is interested in or watching a show or movie will be critical in offering them more relevant content — do they love the sub-genre, the lead actor, the director, the background music, all of this. Deep understanding of user preference and context can lead to more nuanced and creative recommendations like “Family friendly french movies with english sub-titles for your household” or “20 min episodes and quick laughts on weeknights from your favorite series”. For new users, companies should invest in real-time signal ingestion and personalization of content. For example, viewers who click or watch a show or movie in their first session should see improved recommendations on the homepage based on that. With these changes, personalization on streaming platforms should improve leading to higher user engagement and retention.
Pillar 5: Leverage human expertise where it makes sense
Search and AI systems are not always up to date with recent world events. For example, a recent movie or tv show awards ceremony, an upcoming festival, a new election season, a new music festival and more. Media consumption is often connected to local and global events like these and it makes sense that recommendations account for these. For example, building collections like “Movies for summer” or “Movies to watch before the upcoming award season” can delight users and grow engagement. Many streaming companies already do this, though the collections are often generic and not personalized or well targeted. For example, for a user who watches movies in French or Hindi, showing relevant content in that language and potentially from countries that speak those languages, will be more useful. This is where combining human expertise with AI and personalization systems can create delightful effects for customers.
Important considerations
User privacy and controls
We live in a world where privacy matters more every day. It is important for streaming companies to consider and define how they capture and user viewer data. Companies should offer users the ability to control how they capture their usage on their platform and users should be able to opt out. Depending on the company ethos and policies, the default could be different. For users who opt-in, the experience can be more personalized based on their searches, clicks and views while for users who opt-out, streaming companies can offer a wide selection to select from and potentially smart filters to narrow down options quickly. The number of options shown to an opt-in user on their homepage might be limited and more curated compared to the opt-out user. To ensure privacy for all users (opt-in and opt-out), companies can use on-device machine learning to generate recommendations so no user data ever leaves their personal devices and stays safe. In addition, users should be able to control their default language, preferred device and quality to stream on, and default setting for sub-titles.
Cultural and temporal context
As streaming companies improve their content discovery and personalization enginers, it is important these systems account for cultural and temporal context, specially when human oversight is limited. Every country, community or geography has cultural moments and events that shape what people watch, like and enjoy. For example, multiple blockbusters movies and shows are released in India during the festival of Diwali creating a huge jump in searches and views for these new movies and shows. Similarly, after the major award ceremonies in the US in January and February, the award winning shows and movies see growing interest from viewers. In the same manner, sporting events have their moments and timelines. In addition, users also have their habits and moments that vary depending on their personal situation. Families often watch a movie together on Sunday afternoons or allow their kids to engage with video content more during their vacations. Ensuring the systems and interfaces we build understand these cultural nuances and personal preferences will be important to design experiences that customers find intuitive and seamless.
Video streaming services have all started to look and feel the same and there are not many differentiators, except original programming. Netflix is known for it’s largest collection of original content followed by Prime Video, HBO Max, Apple TV, Disney and others. Apple is unique in offering premium content with high-production value while others like Prime Video and Netflix have many hits and award winning series. While Prime Video and Hulu are marketplaces with access to other channels like HBO, Showtime and Discovery, Netflix, Disney and Max keep it limited to owned or original programming only. The services differ a little in pricing, number of devices supported and number of profiles offered. However, the content discovery experience, personalization, streaming quality, and other features like offline viewing are pretty much the same across these services. Viewers often flock and switch between these services, subscribing and unsubscribing, to get access to their favorite content and then leave when they finish. With the right investment in content discovery, however, companies can differentiate themselves with better experiences that are more personalized.
