By Ross Nicol, VP EMEA, Zefr
As the sheer magnitude of the COVID-19 crisis became clear, consumers tuned into news content in their tens of millions to keep track of every latest twist and turn in the pandemic. At the same time, publishers and producers massively lost out on programmatic ad revenue due to basic keyword blocklists for ‘Coronavirus’ and associated terms, even when stories were positive in sentiment.
By now, this tale of wasted opportunity is well known. The industry, however, should still pay attention to its key lesson: there is no one-size-fits-all strategy for content adjacency. Blunt brand safety tools aren’t enough anymore, and to achieve a finely tuned balance of maximum engagement and sensitivity, a more nuanced approach to assessing suitability is needed.
While universally applicable across formats, the need for more focus on suitable content is especially important with video. As one of digital advertising’s fastest-growing channels — and the main driver behind overall growth in 2020 — video is taking an ever-larger portion of ad budgets and is expected to account for 60% of display spend by 2024. To ensure healthy returns on sizeable video investments, getting ad placements right will be essential.
Elusive by nature: the video suitability challenge
Video comes with a unique set of targeting considerations. Research conducted with the IPG Media Lab revealed that while 84% of consumers understand that YouTube ad placements are intentional, just 25% feel that brands did a good job at appearing adjacent to the right content.
Video, by its nature, is highly subjective. Ten people might watch the same video and have varying perceptions of what it is about, which makes it hard to ensure safety and resonance at scale. Additionally, low volumes of text mean contextual stacks that rely on keywords and semantic intelligence alone will have limited capabilities across the video. With fewer signals to guide video content classification than with text-heavy webpages, there is a stronger chance of false positives and negatives, where failure to track outlying indicators results in videos being incorrectly identified as high or low risk. For example, news and video revenue decreased when the term ‘Trump’ became a red flag for brand safety blockers in 2019, regardless of whether the content was politically affiliated or not.
As a result, successful video targeting requires a different architecture than legacy forms of digital media. Brands must harness deeper analysis that goes beyond keywords to establish the suitability of each video, fuel accurate categorization, and open up a wider store of valuable inventory.
Adding a human touch to video algorithms
Overall, keyword-based tools are subject to two core issues. Firstly, terms are too broad and ever-multiplying; often running into lengthy blocklists that restrict advertising reach and don’t allow for the diverse sensitivities of different brands. Secondly, content analysis depends exclusively on keyword present assessment and algorithms, meaning vital subtleties are overlooked. Clearly, the common denominator here is the limited scope and that means the best answer to both of these problems lies with better accommodation of variability.
Video is multi-dimensional, as are individual brand views of suitable media. So, rather than binary analysis and exclusion lists, what’s needed is greater input from the one machine capable of deeply comprehending and adapting to nuance: the human brain.
Adding an additional layer of manual, human review gives smart tech the ability to pinpoint true video context and enable secure, inclusive ad targeting. Using training data generated by human suitability definitions and content evaluation, algorithms can learn how to spot small signals that uncover the meaning and sentiment of videos, instead of leaning solely on text. This knowledge can then be applied to drive large-scale analysis across platforms and direct spending towards content that offers the ideal adjacency for specific brands.
Working towards a more suitable future
Of course, driving an industry-wide transition to versatile video analysis will require more than simply increased human involvement on the vendor side. Advertisers and brands will also need to play their part by setting preferences for their teams to judge success, including refined parameters about what is a suitable and acceptable risk. Fortunately, huge global initiatives are emerging to help them drive a more objective view of brand suitability.
Joint efforts between the American Association of Advertising Agencies (4A’S) and the Global Alliance of Responsible Media (GARM) have produced one set of unified suitability definitions for the industry to follow. Featuring 11 categories — such as terrorism, debated sensitive social issues, or arms and ammunition — the new standard marks a significant milestone in tackling issues with video perception subjectivity and improving transparency for brand advertisers, especially for those running campaigns in black box environments.
Brands now have a shared baseline that can be used to roll out universal categorization of content suitability in line with low, medium and high thresholds, and their own bespoke restrictions. For example, this could see advertisers aiming for low-risk content in the ‘Adult and Explicit Sexual Content’ category serving ads beside sexual education videos, while those operating on higher-risk categories might select a wider range of placements in the context of movie scenes or music videos.
The framework is a massive step in the right direction and provides the foundation for enhancing common understanding across the ecosystem, as well as giving brands the means to ensure ad spend is invested in appropriate and ethical content.
As the industry shifts from brand safety to brand suitability, it requires a new mindset and new tools to solve for video. Already losing its value for general display, mass blocking certainly isn’t the smart choice for scaled video platforms. Simultaneously optimising safety and reach will mean moving away from basic keyword methods and embracing a more human approach. By switching to analysis led by human cognition alongside machine learning, brands can create comprehensive tech stacks that not only reduce risk, but also unlock opportunities to tap into suitable and powerful video content.