Can ML-powered ad-targeting solutions offer notable advantages over current algorithm-based systems? John Maxwell Hobbs reports.

Broadcasters are keeping a close watch on recent advances in AI and machine learning (ML) technologies and their use in areas such as Dynamic Ad Insertion (DAI). These advanced technologies have the potential to reshape the way advertisements are delivered and consumed, promising enhanced relevance, efficiency, and revenue for both broadcasters and advertisers.

Dalle] A photorealistic image of a modern broadcast studio showcasing advanced d

A futuristic broadcast studio showcasing advanced DAI using AI

Source: DALL-E/John M Hobbs

Current Use of AI/ML in Dynamic Ad Insertion

AI and ML are already making their mark on DAI by facilitating real-time decision-making and sophisticated targeted advertising. According to Daniel Pike, the Chief Product Officer at Covatic: “AI/ML algorithms analyse vast amounts of data, including user behaviour, preferences, and context like time of day or content consumption patterns, to dynamically select the most relevant ad for each impression,” he says. “These technologies can predict which ads will perform best based on historical engagement metrics, thus enhancing ad relevance and effectiveness.”

Daniel Pike 2

Daniel Pike, Covatic

AI and ML have the potential go beyond basic demographic targeting by leveraging a large number of data points to create a comprehensive understanding of individual users. This includes factors such as browsing history, social media interactions, and even real-time geographic location. “Such models also bring with them several practical, legal and ethical challenges,” warns Pike. “One challenging area is user privacy,” he says. “Distributed AI and ML offer significant privacy advantages over traditional approaches because personal data can stay with individual users and doesn’t need to be pooled in a big data lake somewhere.”

Read more How can streamers maximise DAI and avoid the complexities of GDPR?

Johan Bolin, Chief Business Officer at Agile Content, points out that current systems are already using forms of AI but have the potential to become more sophisticated. “AI has a very broad definition,” he says. “A machine matching a user profile with related advertising is well established. Next up would be to track the efficiency of the ads and feed that back to the ad decision server to ‘learn’ which ads work best in which circumstances. While part of this is ‘private magic’ inside various ad insertion systems, I’d argue that Google, Amazon and Facebook already have this to various extents.”

Effectiveness Compared to Traditional Algorithms

There are plenty of effective algorithm-based ad-targeting solutions available today. The question is whether AI/ML-based targeting can offer notable advantages over the current approach. Pike elaborates: “AI/ML-based targeting can be more effective than traditional algorithms due to its ability to process and interpret massive datasets rapidly and accurately. Traditional methods often rely on static, rule-based systems that can miss nuanced user behaviours and preferences. In contrast, AI/ML continuously learns and evolves,” he says.

Bolin concurs: “Exactly how effective it is depends on how it’s applied, but with enough data, and enough ads, we can assume this is very effective,” he says. “Given that the core business model of several of the tech giants is advertising, it’s reasonable to assume that significant parts of the huge budgets assigned to AI are actually motivated by increased revenues from better advertising.”

Addressing the Black Box Problem

One of the critical challenges with AI/ML in DAI is the so-called ‘black box’ problem – in many cases, even the developers of the AI technology do not understand how their systems achieve specific results. This lack of transparency can make it difficult for advertisers to understand why certain ads are selected over others and to ensure that the AI system is functioning correctly.

Pike underscores the importance of transparency and accountability in AI systems: “AI solutions can sometimes be difficult to fully understand, conceptualise and explain,” he says. “There is always the risk that AI is filling a capability gap and providing nonsense answers - i.e. a clever AI is providing plausible but bogus outcomes. If you don’t know how the AI has come to the decision it has, how can you tell a good AI or ML solution from a bad one? A reasonable starting point is to look at the inputs to the system. If you are seeking to understand the age, gender or purchasing preferences of a user, for example, but you look at the input data and can’t see any plausible way an informed person with lots of time and energy could ever reach a sensible decision based on that data, then it is probably fair to say an AI can’t either. If you have a truth set and can, test the system’s performance thoroughly: test, test and test again.”

Compute and Energy Implications

The powerful capabilities of AI/ML come with considerable compute and energy demands. Training and running AI models, especially deep learning algorithms, require substantial computational resources. Traditional algorithms, being less complex, consume less energy.

Johan Bolin

Johan Bolin, Agile Content

Bolin points out that determining the resource footprints of these sorts of systems “is a little like asking ‘how long is a rope.’” He points out that it all depends on what the AI is being asked to accomplish. “In a rather basic form, it’s just matching a user profile with matching advertising,” he says. “If you extrapolate though and imagine that in the future the entire ad is created ‘on-demand’ and the user targeted with generative AI, and the advertising is on-the-fly encoded into the video as a product placement, this will consume much more energy both in the AI creating it, as well the encoding for every stream and ad.”

The challenge for the industry is to balance the efficiency gains from AI/ML with the need for sustainable energy consumption. Edge computing, which involves processing data closer to where it is generated rather than in centralised data centres, offers a promising solution.

Pike explains: “Edge computing and specialised AI chips are helping to mitigate the compute and energy demands of AI/ML processes by optimising performance and energy efficiency. Distributed AI is definitely an exciting new frontier,” he says.

Read more DAI and Sustainability: Session-Based Solutions and Efficiency in Ad Delivery

Revenue Potential for Broadcasters and Advertisers

Pike believes that AI/ML-based targeting has the potential to significantly boost revenue for broadcasters and advertisers. By delivering more relevant and engaging ads, these technologies can make ad inventory more valuable. “AI/ML-based targeting holds substantial potential to significantly enhance revenue for broadcasters and advertisers,” he says. By delivering more relevant and engaging ads, AI/ML can drive higher engagement and conversion rates. This precision targeting means ad inventory is used more efficiently, potentially increasing the value of each impression. While initial gains might be incremental as systems learn and optimise, the long-term benefits are likely to be substantial.”

Bolin also believes the use of AI has strong potential for increased revenue by changing the nature of the way products are advertised. “Especially in the more advanced, longer-term implementations of Gen-AI,” he says. He can envision the technology being used for ad creation or product placement. “Possibly by extending some scenes with ad scenes and sequences rather than traditional breaks,” he says. “This would be a way to get much more ad load into the content without increasing the intrusion of the content. This way it provides value to all parties since the experience is enhanced at the same time as the inventory size and value is increased.”

Trust and transparency

Generative has the potential to create real-time ads unique to specific viewers by leveraging data on user preferences and behaviours to make them more relevant and engaging.

Dalle A photorealistic image of a modern broadcast studio showcasing advance

AI has the potential to power interactive ads that adapt to user interactions in real-time, with a futuristic broadcast studio using AI

Source: DALL-E/John M Hobbs

However, this also raises concerns about intrusiveness and user trust. Pike stresses the careful balance required between personalisation and privacy, ensuring transparency and user consent to avoid negative perceptions. “Generative AI holds promise for creating highly personalised ad content, but it walks a fine line between relevance and intrusiveness,” he says. “On one hand, generative AI can craft unique, personalised ad experiences in real-time, which could significantly enhance user engagement and effectiveness. On the other hand, if not handled with care and transparency, such personalised ads could be perceived as creepy or invasive, potentially damaging user trust. It’s imperative for the industry to balance personalisation with user privacy and consent.”

Read more SMPTE: Transparency and education the keys to managing AI

Bolin echoes this sentiment: “In the beginning, it will feel very intrusive and there will be pushback. As this technology gets even better, it might be that users don’t even perceive the ads as ads - but that needs a re-definition of TV ads.”

The Broader Impact of AI/ML on Advertising

AI/ML has the potential to redefine the landscape of advertising in significant ways. Beyond personalised targeting, these technologies can enable more dynamic and immersive ad experiences. For instance, AI can power interactive ads that adapt to user interactions in real-time, or augmented reality (AR) and virtual reality (VR) experiences that create highly engaging brand interactions.

“AI can power interactive ads that adapt to user interactions in real-time,” says Pike. “Additionally, AI-driven analytics provide deeper insights into campaign performance, enabling real-time optimisations and more strategic decision-making. As AI/ML continues to evolve, we will witness a shift towards more intelligent, engaging, and effective advertising.”

Bolin agrees, stating: “In the ultimate form, AI/ML can erase the idea of ‘ad breaks’ by replacing them entirely with content-integrated ads.”

The integration of AI and ML into DAI has the potential to deliver a transformative advancement in advertising, but there are also several unanswered questions around transparency, energy consumption, and user privacy. As AI/ML continues to evolve, the industry must navigate these challenges carefully to fully realise the benefits of intelligent, engaging, and effective advertising while deploying the technology responsibly and transparently.

Read more IBC2024 Primed to Lead Innovation, Explore Trends and Foster Collaboration across the Global Media Technology Community