Contextual specialists are usually the first to say contextual targeting is ready to replace third-party cookie for targeting purposes. But Rich Raddon, co-founder and CEO of brand suitability business Zefr, says context will play more of a supporting role to audience-based targeting, which isn’t going away.
In this interview Raddon also discusses the limits of semantic-based brand suitability assessment, and the impact of the WFA’s GARM standards for harmful content.
Zefr uses a ‘Human-in-the-Loop’ system for gauging brand suitability. How do you manage human review of video content at scale?
We’ve invested heavily in finding the right balance between human cognition and machine learning – as with the scale of platforms, it’s impossible to have a human review for every piece of content. Our data science team works with our in-house moderators, and we have built a process that integrates into crowdsource platforms to generate scaled reviews. But in order for the human reviews to be valuable, there needs to be intense quality control and objective definitions for humans to be able to understand at scale, and then advanced machine learning and AI to actually put that training data to work. Our focus is the harmony of those components to make brand suitability much more precise for video, and the work is never done.
What are the limits of semantic-based brand safety/suitability assessment?
Semantic-based technologies work well when there is a lot of text to assess, as the technologies were built for the open-web. On video platforms, where text is limited, it leads to an incredible amount of false positives due to mislabelling. It ends up leading to massive over-blocking of content, and misclassification of “risk” in verification reports.
With the emergence of GARM standards, we’re finding that human cognition is a much more accurate way to capture brand suitability in video than semantics are due to this very issue.
How granular can your categories get with your ‘human-in-the-loop’ methodology?
We’re focused less on granularity, and more on precision to make sure we’re providing accurate and quality data that scales for our brand partners. Our current focus is using industry standard categories – the GARM Brand Suitability categories of Low/Medium/High risk – to build off of industry guidance. We think the industry is better served with data aligned to common standards, rather than having to sort through each vendor’s custom taxonomy.
How do you assess the opportunity for contextual/content-based targeting in the post-cookie era? Do you think it can completely compensate for cookie-based audience and behavioural targeting, without any change in performance?
Audience based targeting is not going away. We think context plays a complementary role in audience targeting – particularly in reducing waste from shared Device IDs. Contextual is not a silver bullet, but it has its place in the marketing funnel.
Is contextual targeting completely immune to privacy concerns? Or does contextual targeting have to be run a certain way to be privacy safe?
Context certainly is a solve for many brands concerned about privacy, as the data is about content, not PII. The important takeaway for the industry should be about using data in a more effective way that embraces consumer trust – not trying to recreate the old problems with band-aids.
Zefr has partnered with YouTube to measure brand safety and suitability against GARM standards. What is special about these GARM standards and what do they mean?
The GARM standards are absolutely critical and we believe represent the next era of media and marketing responsibility. Before GARM, brand suitability was entirely subjective and each company would be able to create its own version, leading to an incredible amount of confusion, and ultimately devaluing the space. With GARM, there are consistent definitions that are being agreed upon in the industry, across brands, agency leadership, and platforms. With these standards, the industry can become more consistent and measurable in brand suitability – ultimately benefiting all parties.
When it comes to brand safety in regards to influencers, is it ever really possible for brands to be fully assured of brand safety? For example, with the recent allegations relating to YouTuber David Dobrik – was it possible for brands to tell it was a risky channel prior to the allegations breaking?
As the GARM standards demonstrate, there’s no such thing as “no risk” on platforms. But the progress that platforms have made in the Brand Safety Floor by removing, demonetising, and reporting on violations, has been tremendous. The next era will see the industry move beyond the Brand Safety Floor – and we believe brands can take a proactive approach by aligning with the GARM standards.