Over the last few years, machines have gotten a lot wrong. From safety to privacy to context, machines without proper focus have proven they aren’t quite ready for prime time. As a result, the industry adapted — leveraging old tools to solve new problems.
In traditional media, like TV, all of the content is knowable; networks have every script, schedule, and program approved in advance, and brands can control where their spots run accordingly. And then platforms arrive.
The best we can do is whitelisting/blacklisting, which only offer snapshots of content at a fixed moment in time to mitigate the dangers of a dynamic content environment.
How does the industry solve this incompatibility?
Even the savviest keyword algorithm can’t capture cultural nuance before it happens. For machine learning to drive effective targeting on dynamic platforms, machines need to be continuously monitored and trained on behalf of brands to understand these ever-evolving elements and adjust in real-time with brand preferences in mind.
The winning formula is the brand-focused human application of nuance into the rigid patterns of machine learning, which optimizes the outcomes that machine learning manages.
This is leaps and bounds beyond the reactive, manual keyword whitelist/blacklist adjustments after damage is done.
So how can human supervised machine learning processes put an end to keyword-based whitelist and blacklist strategies?
Eliminate False Positives in Context
In search and display advertising, keywords are the primary drivers of contextual alignment. But in video, keywords alone can’t properly capture the intent of the video. Human-supervised machine learning separates the signal from the noise by actively finding patterns across thousands of video attributes.
Even the most sophisticated keyword process will miss nuance. The keyword “Eagle” could be the band, the 2018 Super Bowl Champions, a wildlife video, or a new dance trend. Additionally, a new concept like “Drill music,” which has been linked to violence, can’t be predicted via keywords, much to the chagrin of brands found adjacent to it.
Humans tease out difficult patterns and teach machines to be more robust than a static keyword list could ever be.
YouTube, the world’s largest video platform, has content that is relentlessly dynamic. There is too much content for humans to monitor every video, while there is too much nuance for unfocused algorithms to do it alone.
Take, for instance, a brand that aligns with cooking, health and wellness, and travel content, while absolutely steering away from Non-English and kids’ content. With this specific and intricate set of parameters, designing a single set of keywords that is both applicable now and continues to be applicable in the future is a near impossibility. Maintaining whitelists is futile.
If a news story breaks that a certain type of food is being recalled by the FDA, the advertiser would naturally want to remove itself from any content that features that particular food item. But by the time the distinction is incorporated into a blacklist, the ad may have already landed adjacent to the radioactive content.
This is where human-supervised machine learning becomes critical.
Qualified reviewers can proactively and continuously watch videos to determine if each video matches the brand’s content guidelines. These consensus reviews become the ground truth, training data in determining whether a video is campaign-worthy.
The data—both positive and negative—will serve to train each custom machine-learning model to dynamically identify more videos that match the brand’s preferences, while excluding ones that don’t.
Rather than reinventing the wheel and constantly adding and removing new keywords, this structured approach can become your brand’s dynamic content targeting strategy. In simple terms, you are moving from a reactive to a proactive process, combining the complementary strengths of human beings and machines.
By leveraging human-supervised machine learning, marketers can increase their precision against relevant content, while improving workflow efficiency and enhancing consumer trust. It has also shown to drive purchase and recommendation intent, and encourages consumers to watch rather than skip ads.
2019 will be a year filled with advances in machine learning from some of the biggest tech companies in the world. It will help marketers with brand safety, as well as with ad performance on KPIs, such as viewability and efficiency.
But the true promise for brands requires human cognition and machine learning to find patterns that build your brand, not just protect it, on the biggest platforms in the world. Pointed the right way, machine learning can make your brand’s advertising more powerful than ever.
Machines alone are never going to think like CMOs -- but it’s our job to train them to.