Peter Wallace, EMEA Managing Director at GumGum, explores how AI tech is helping advertisers bring far more contextual accuracy to the way they target video ads
The online video advertising market is on a powerful trajectory of growth right now. eMarketer predicts that in the UK, the sector will jump 15% to £3.72bn, strengthened by the leap in YouTube viewing throughout lockdown. Globally, the picture is similar.
Given this injection of budget, it is timely that the methods by which we target and anti-target (i.e. maintain brand safety) video ad campaigns are maturing. Artificial intelligence is bringing some much-needed nuance to the science of working out where video ads should appear, helping advertisers earn more effective consumer attention, and publishers deliver more valuable supply.
Frankly, the changes aren't before time. For too long, we have relied on analysing simple metadata to help us understand the nature of video content. Numerous brand safety stories in video over the years highlight how unreliable this type of analysis is. And as third-party cookies take their final bow, audience data becomes an even less scalable solution for video targeting.
Thankfully, we’re now seeing a resurgence in contextual targeting, which is negating the need for cookies, along with increased capabilities allowing advertisers to understand the true sentiment of the content; making this a scalable, brand safe solution. To apply this to video, we need to move beyond basic text only analysis - welcome to the age of content level targeting.
Content targeting concerns itself with the development and integration of two key areas of artificial intelligence - natural language processing (NLP) and computer vision (CV).
NLP allows computers to not just analyse text but to understand the broader context and meaning of words. For video, this involves analysing the voice-over or audio track and the language that it contains. Take the word ‘sex’. If this is mentioned in the audio, it could suggest the content is sexual and inappropriate for most advertising. But it could be harmless - for example, if the word is referring to someone’s gender. NLP can analyse words at this deeper level and understand the nuance of language.
Computer vision brings the same level of intelligence to image-based content. The tech allows computers to analyse huge volumes of video content online, frame by frame, and understand every object and scene. CV could, for example, scan a video, identify the green grass of a pitch, advertising hoardings at the side and the faces of well-known players, and automatically deduce that this video is of a Premier League football match - a perfect piece of content for a sports brand.
This subtlety of approach improves the targeting of ads but also the accuracy of anti-targeting – avoiding video content that could be dangerous or just unsuitable for an ad for a particular brand to appear alongside. Through CV, a computer could be trained to spot an object within a video such as a gun or a knife, or it could identify offensive symbols like a swastika.
Fears over ads appearing next to controversial or dangerous videos has led many advertisers to adopt ever more draconian ways of avoiding this content, to the extent that vast amounts of perfectly safe inventory is lost to them. Through these advances in AI, we can refine the identification process and evolve a system that allows us to stay away from difficult content, but still identify the kind of quality placements for an effective video advertising campaign.
In an effort to make these technologies even more available, we at GumGum have established a partnership with IRIS.TV, a cutting edge tech platform that analyses and classifies video content broadcast on publisher sites owned by companies including News Corp and CBS. This then makes it easier to offer viewers other footage they might be interested in and for advertisers to cue up video ads relevant to the context of the user's viewing.
Our NLP and CV technologies have taken IRIS.TV’s intelligence to the next level and offered advertisers a new, advanced tool for safely and accurately reaching vast video audiences without encroaching on their data privacy.
This latter point will only become more relevant. It was only relatively recent that behavioral targeting, reliant on ever-more intrusive levels of data capture about users' browsing activity, was seen as the bedrock of digital marketing. GDPR has curtailed that, with progressive marketers now keen to find contextual solutions that allow advertisers to match up ads with appropriate content at scale. The pioneering work that’s going into the analysis of both the sound and vision of video means advertisers will increasingly have the tools they need to take full advantage.
Posted on: Wednesday 19 August 2020
