In association with Illuma.
In this exclusive byline, Wieger Holvoet, programmatic technical lead GAM team at Dentsu, outlines how applying contextual-AI can help marketers reach campaign-relevant audiences in CTV, highlighting Dentsu's collaboration with Illuma.
It could be argued that Connected TV (CTV) is the most exciting new frontier in digital advertising since the dawn of programmatic. It’s one of the top priorities for many of our clients, and the true scale of this media opportunity is only just beginning to be understood. But behind the scenes, the ad tech infrastructure is still catching up with the pace of the audience switch, leaving many marketers unsure about how to translate their open web targeting practices into CTV.
As things stand, the audience addressability that we’ve taken for granted on the open web is not directly transferable to CTV environments, and of course, with the demise of cookies, will soon be impossible to scale. The basis of CTV targeting is currently generic demographic, topic, and geographic data, with the obvious downside being that these parameters are quite broad and certainly not campaign- or brand-specific in any way. So, in a marketplace where audience reach, relevance and addressability are so important, how can we target in a more granular, audience-specific way in CTV, without encountering challenges of scale?
In recent months, the Dentsu Global Addressable Media team has thought deeply about the important question of establishing audience relevance in CTV — how to find the right viewers for our campaigns with the limited data available — and realised that it’s possible to push further. So we went back to the basic premise of audience targeting being based on a person’s content consumption — ‘behavioural targeting’ — but applied it laterally, powered by AI, to import live insights into CTV from other platforms. Let’s explain what that looks like.
Same audiences, different places
In the advertising industry, we’ve long understood that to reach audiences successfully, targeting needs to be omnichannel. Audiences and their interest in a brand’s advertising are very rarely, if ever, defined by a singular platform. This person uses social media and nothing else; this person watches TV on a connected device but never goes online to browse the web – these scenarios are inconceivable. Relevant audiences can be found at different times on different and multiple platforms; online behaviour is fluid.
With this typical cross-platform behaviour in mind, and knowing what we do about behavioural targeting at Dentsu, we started to explore ways to capture real-time contextual information from the browsing of highly valuable first-party audiences on the open web, where this behaviour is both visible and supported by multi-dimensional signals. We then modelled it out to make audience-driven contextual targeting decisions in CTV, extending the behavioural insights cross-platform.
The key difference between this approach and existing behavioural targeting is the use of live contextual information as the qualifier of relevance. Traditional behavioural segmentation methods have required cookies in order to track a user’s past behaviour, generally over a 30-day lookback window. This more contemporary way of working, which takes advantage of machine-learning, can track and model content consumption behaviour in real-time, and marry this with live KPI campaign-outcome data; doubling-down on audience relevance and also intent in a way that’s not been possible before.
The real step-change happens when we take these behavioural insights and simultaneously apply them into environments where audience data is sparse, such as CTV, enabling audience-driven contextual targeting built on the live behaviours of our first-party audience.
Tuning into live engagement signals
Fortunately, the expertise to do this already exists – Dentsu turned to contextual-AI specialist Illuma, where the team have spent the past year developing its long-established audience-expansion capability so that it can work across platforms. The resulting AI-engine now takes live, rich contextual inputs from the web, and transposes them into CTV recommendations; it’s a capability that can also assess audio and visual from CTV files themselves, to complete the loop.
The resulting use-case study suggests we now have a dynamic, fluid, campaign-specific way of making CTV impressions addressable, by matching the cross-platform behaviour of audiences. And, on indexes where we can track media quality, this methodology is now delivering even stronger results.
This live, audience-driven targeting also generates a deeper understanding of viewer intent and behaviour, allowing for insight-driven targeting in CTV as opposed to off-the-shelf generic solutions. What is more, as the contextual environments being targeted are fluid and change based on live behaviour, the campaign can scale without losing relevance. This type of recommendation engine, overlaid with true user behaviour, works to solve real world problems and shows AI at its very best.
There probably won’t be a single silver bullet to solve buy-side addressability in CTV (certainly not one that’s arriving imminently), so achieving any step forward is very positive for both advertisers and media-owners alike, and for our clients, it’s already proving to be game-changing.
We worked exclusively with Warner Bros. Discovery’s streaming service discovery+ to access premium supply; they also found this a beneficial way to work.
Alex Hodge, head of digital ad sales and innovation at Warner Bros. Discovery, said:
"Streaming platforms such as discovery+ offer an exciting new opportunity for advertisers, but in order for us to really leverage the benefits, it’s important we continue to work together as an industry on innovation. A key challenge is finding the right targeting solutions to support advertisers in reaching the right audiences in the highest quality environments, and so by using audience signals, this trial campaign opened-up a broad range of streaming inventory which could be intelligently scaled, something which is vital for advertisers and streaming platforms alike."
As advertising in linear and CTV continue to merge, alongside closer collaboration between buyers and sellers, we should see more of this digitally native capability being enabled by smart ad tech.
Meanwhile, other developments are taking place in the space, with OpenRTB 2.6 introducing the concept of ad-podding, allowing buyers to bid on specific pods of CTV content. Channel 4 and Nectar 360 are also collaborating with Infosum on data bunkers, to make CTV more addressable on a user basis.
It’s all an excellent indication of how the industry is collectively maturing so that everyone can benefit from the opportunities of this new frontier. The future of CTV targeting is very bright indeed.
Wieger Holvoet is the technical lead in Dentsu’s global addressable media team (GAM) of programmatic traders, where he focuses on innovative programmatic advertising strategies with ad tech partners for global clients.