Making it personal: Wowza explores the role of AI in sports audience engagement and monetisation

By Barry Owen, chief solutions architect at Wowza.

Artificial intelligence (AI) has been on the tip of a lot of tongues over the past several years, largely thanks to the advent of generative AI. However, the underlying tools and models that make things like generative AI possible also serve other more behind-the-scenes purposes. Here are some interesting ways in which AI is being put to work in the video streaming industry, particularly sports streaming, to improve engagement and expand monetisation opportunities.

Personalised recommendations and playlists

Have you ever wondered where that list of recommended titles on your Netflix account comes from? Netflix uses predictive models that cross-reference viewer behaviours and ratings and predict how they are likely to respond to specific content. When they say that rating content you’ve watched will help them formulate better recommendations, they aren’t kidding.

Lately, Netflix has been taking this automated approach to recommendations a step further. It uses personalised artwork on thumbnails, using machine learning (ML) to select frames from shows that might appeal the most to individual viewers. This also extends to ‘dynamic sizzles’, which are essentially brief personalised trailers. Netflix achieves all of this via contextual bandits – ML algorithms that take context such as user behaviours, preferences, and demographics into account.

Netflix is a prime example but it’s hardly the only streaming platform to take advantage of these tools. Wherever these features reside, it’s common practice to use increasingly nuanced algorithms to power them, including sports streaming platforms with VOD.

Screen quality optimisation

Many VOD streaming companies, sports and otherwise, use both automated and manual inspections of streaming data before and after encoding. These automated inspections rely on ML to learn from past manual inspections and predict how well a stream is likely to perform, applying a pass or fail to it.

Streaming platforms also apply deep learning to improve stream quality beyond what adaptive bitrate streaming (ABR) can accomplish. This method trains computers to recognise certain types of video content and make decisions based on that recognition. While ABR will adjust video bitrates in response to feedback about end user bandwidth, this deep learning method adjusts video bitrate in response to what’s happening in the video itself. For example, a sports live stream may benefit from larger or smaller bitrates, depending on the level of action in any given moment.

Real time translations

Real time translations will add closed captioning in several languages to a live stream. Automating this process can make the captions faster, more accurate, and available in more languages. Companies have been exploring a form of ML called Automatic Speech Recognition (ASR) to improve speed, accuracy, and scope of available languages. Historically, ASR has been spotty, particularly where dialects are concerned, but it’s improved dramatically in more recent years.

Possibly most impressive is AI’s ability to identify and accurately interpret idiomatic expressions. These tools look for patterns in phrases and context to infer meaning. They can then apply that learning to new scenarios. Some of these tools also let you proactively train them using glossaries and style guides for a more targeted approach. These translation and captioning tools are particularly handy for live sports streams given their global appeal.

Real time highlights and auto-clipping

Just as AI can help optimise playback quality by intelligently identifying certain types of video content, so too can it single out specific moments for highlighting in real time. This is especially handy for live events where clips and highlights are particularly relevant.

In these cases, the AI tool (sometimes embedded in the camera itself) not only identifies moments worthy of highlighting, but also automatically creates a highlight for immediate use. That same technology is used to auto-clip for the purposes of social media posts, sizzle reels on YouTube, and more.

Targeted advertising

This one might sound like old hat, after all the idea that the ads we’re shown on a website or via YouTube are not purely random but based on an algorithm is nothing new. However, with the more recent explosion of effective AI technology, our ability to categorise people and predict their wants has become much more nuanced.

Programmatic advertising gathers data in real time, cross-referencing audience demographics, behaviours, and context to optimise the purchasing and placement of ads. Relevancy increases the value of ads for those advertising and is less annoying for viewers.

AI is also being used to find not only the right viewers, but also the ideal spaces within a piece of content for advertising. If you’ve ever been watching a show on Hulu only to have the mood abruptly broken by a loud and bright commercial, then you understand the importance of choosing the right moments and matching tone. AI can help identify good places for breaks and choose ads that complement the overall viewing experience.

Analytics for audience behaviours and content performance

Analytics have long been an essential part of the video streaming workflow. A robust analytics package can tell you about both stream health and audience behaviours. More specifically, it can tell you who is watching what type of content, from where, on what type of device, and for how long. It can also tell you how well the stream performed. What was the average bitrate? Was there any buffering?

AI has not always been a part of this process, but it certainly has found a home here. As AI becomes smarter and more nuanced, it’s easier to train it to scour and cross-reference large datasets. It can learn to identify key relationships and alert platform administrators of insights.

What’s next?

Our realisation of AI is more of a marathon than a sprint. Everything we see possible today is built on the back of numerous efforts to make machines and associated learning models smarter. The more we find useful ways to teach it and apply it, the more we’ll be able to do with it in future. So, what’s next for AI in sports streaming? We can’t wait to find out.