Project managers Lisa Kimura and Prayushi Amin share key findings from our latest media trial, Solving Brand Suitability.
Project managers Lisa Kimura and Prayushi Amin share key findings from our latest media trial, Solving Brand Suitability.
By SHOSHANA WODINSKY. Published by ADWEEK on 17 October 2019.
When it comes to brand safety, brands might be better off relying on humans, rather than machines.
According to new research out today co-produced by L.A.-based contextual data company Zefr and the advertising analytics outfit Magna, media buyers could get better bang for their buck by including humans in their content-review process, rather than relying on preset white- or blacklists. Zefr—which patented its “human in the loop” brand-suitability algorithm that guides machine learning models using human review—found that the resulting ads turned out to be more relevant, reached consumers that were more likely to be in-market and were more likely to convert as a result.
“In the last two years, the platforms have gotten really good at machine learning for brand safety,” said Andrew Serby, Zefr’s vp of marketing. “All the content that’s incredibly problematic in terms of the obvious stuff —whether that’s violence or hate speech or crime, they’ve done a really good job.”
What they aren’t good at is nuance.
Efforts to keep brand-safety snafus at bay have seen legitimate news sites and niche publishers fall onto blacklists for publishing kosher content with nonkosher keywords. According to Serby, this is the reason why Zefr is a company that’s trying to shift the advertising zeitgeist from conversations of the black-and-white world of brand safety and into the shades of grey of “brand suitability”—case-by-case definitions of what a particular brand is comfortable advertising against.
Unsurprisingly, humans are better at picking apart these nuances than their machine counterparts, which is why Zefr’s tech works on a brand-by-brand basis, taking their inputs about the kind of content they’re comfortable appearing alongside and building machine learning models off of that.
“Instead of targeting what you don’t want to be around, it’s about being proactive,” Serby explained. “It’s about deciding the kind of content you do want to be around, and finding the content that complements your brand’s message and drives better results.”
For the study, Zefr worked with three brands in three verticals: Nationwide, Ubisoft and ScottsMiracle-Gro, and targeted YouTube ads from these companies against more than 3,000 YouTube viewers who were targeted in one of four ways. The first reflected the typical buy, targeting popular channels with an eye toward targeting a certain demographic. The second was a channel-by-channel buy and the third was based on keywords that were brand-relevant. The fourth used Zefr’s human-centric algorithm, using “signals” from each of the three brands to determine which kinds of content would be suitable—and safe—for each.
“This is content that you might not get in trouble for advertising against, but it’s not the content that an advertiser would be excited to promote,” Serby said. “It’s a much more nuanced discussion.”
That nuance pays off.
When it comes to reaching in-market consumers, Zefr’s proprietary methods were found to convert roughly 11% of the time. While that might not sound like a large number, keyword targeting resulted in only 6% of those same conversions, and channel- or demographic-level targeting barely scraped 1% each. All three of those methods reached an in-market less than Zefr’s proprietary tech, as well, at 75% of the time, rather than upwards of 80%, as Zefr found.
“When brands determine the signals used to identify content that makes the most sense for them, misalignment between content and ad is curbed, and each ad works to its full potential,” Zefr wrote in a statement.
Aside from the relevancy of an ad in the content it’s playing alongside, there’s also the question of the “quality” of content a brand is willing to appear alongside.
“Typically, a brand will have some definition for quality content, in terms of the premium nature of that content—it’s studio-produced; it’s official, TV-like content,” said Serby. Because Zefr focuses more on brand-led signals, rather than quality, it expands that definition. And the more advertisers expand that idea, “the more they can capture what a consumer thinks is quality on platforms that don’t necessarily look like television,” he said.
The Study, Conducted by Magna and Zefr Explores The Impact of Machine Learning Combined with Human Supervision on Brand Suitability
LOS ANGELES – October 17, 2018 – “Human in the loop” contextual targeting – which uses brand preferences to power machine learning that is overseen by humans – is dramatically more effective than traditional modes of ad targeting, according to “Solving Brand Suitability,” a new study by MAGNA and the IPG Media Lab conducted with Zefr, the Contextual DMP™ for brands and agencies.
The study aimed to provide a foundational understanding of how brands can better achieve “brand suitability,” defined by advertisers as their unique positive and negative contextual preferences. Advertisers are increasingly focused on how different targeting methods fare in achieving it.
The study found that just 25% of consumers think brands are doing a good job of advertising on YouTube and only 18% of those who expect relevance between the ads and the video said that the ads are typically aligned with the videos they are watching. The study then explored different methods marketers could employ to improve “Brand Suitability.” Nearly 4,000 consumers were surveyed on their reactions to ads from three brands across verticals – Nationwide (insurance), Ubisoft (gaming) and Scotts Miracle Grow (paper/manufacturing). Ads were delivered to consumers via different forms of targeting: demographic; channel; keyword; and “human in the loop contextual targeting” (where a team consistently reviews videos in order to train machine learning models). The study revealed that ads delivered through “human in the loop” contextual targeting outperformed all other methodologies in a number of key metrics:
“Contextual targeting is highly nuanced for each brand, especially in video, and traditional methodologies like static ‘whitelists’ and channel targeting often miss the mark, negatively impacting reach and wasting valuable media dollars,” said Rich Raddon, co-CEO of ZEFR. “This study provides valuable industry insights on how brands can take control with human-in-the-loop contextual targeting and increase the impact on every part of the funnel, from in-market reach to purchase intent.”
A somewhat unexpected insight revealed in the study is the considerable opportunity for advertisers to reach audiences by expanding their definitions of “quality” video. 44% of content machines identify as low-quality is perceived as high-quality by consumers who view it as enjoyable and interesting. The study shows that in video, quality is often in the eye of the beholder, and brands can succeed by tapping into this significant pool of largely uncharted, brand-suitable ad inventory.
UM’s Chief Digital and Innovation Officer, Joshua Lowcock said. “This is a firm reminder context matters as much as the data used to find an audience. The more aligned the ad is with content, the more likely consumers are to view the brand as innovative, savvy, trustworthy and one for which they will pay more. Using human-supervised machine learning to help find suitable content is one way of finding that balance.”
Zefr is a contextual technology company that delivers precise and effective contextual solutions for brands and agencies. Its Contextual DMP™ is an identity-less solution that enables brands and agencies to capture, organize, and activate their nuanced contextual preferences at scale, for video and beyond. By leveraging proprietary Human-in-the-Loop technology, the company builds customized and nuanced contextual solutions for major national brands and advertising agencies. The company is headquartered in Los Angeles, California, with offices in New York, Chicago, Toronto and London. For more information, go to: http://zefr.com
About IPG Media Lab
Part of the Interpublic network, the IPG Media Lab identifies and researches innovations and trends that will change the media landscape and how brands engage with their audiences. Since 2006, the Lab has worked with our clients and with industry partners who can help them best adapt to disruptive change. Its expertise, resources and consulting services also help to inform the learnings, strategies and business outcomes of all Interpublic agencies. For more information, please visit www.ipglab.com or follow @ipglab.
MAGNA is the centralized IPG Mediabrands resource for market intelligence, media investment and innovation strategies. The agency utilizes its insights, forecasts and strategic relationships to provide clients with a distinct marketplace advantage.
MAGNA harnesses the aggregate power of all IPG media investments to drive maximum value for its clients through preferred pricing and premium inventory. The agency’s Investment and Innovation teams architect go-to-market media strategies across all channels including linear television, print, digital, programmatic and emerging media. MAGNA is a leader in generating data and technology-enabled solutions that drive optimal client performance and business results.
The agency’s Intelligence unit has been a coveted source of crucial industry information, including media value predictions, for more than 60 years. It produces more than 40 annual reports on audience trends, media spend and market demand as well as ad effectiveness. For more information, please visit https://magnaglobal.com/.
Research points to factors that drive our willingness to see ad before we’re exposed to it.
Mood, needs state and situation are all key indicators of ad receptivity, according to a new study by MAGNA and IPG Media Lab in partnership with Pandora.
The research, which surveyed more than 2,000 respondents who kept an online diary of their digital audio and video consumption over 24 hours, found that audio listeners and video viewers are similar in that they are most receptive to ads when in a state of excitement and when spending time with family.
However, they exhibit distinct differences as well.
Listeners are generally 35 percent more open to ads and specifically receptive when relaxed and focused, while viewers are more easily targeted when they’re stressed.
“This study validates that digital audio and video have infinite moments to reach people willing to be exposed to advertising,” said Keri Degroote, SVP research and analytics at Pandora. “Marketers who understand the when, the where and the what to meet people’s state of mind and needs state will win their attention.”
The study unearthed a number of other important insights.
Ad receptivity is dependent on a person’s emotional state. Good moods (excited, relaxed, focused, happy) means more willing to see advertising for audio, whereas an excited mood translates to better ad receptivity for video viewers. When people are tired, they’re least receptive to ads on both audio and video.
Gen Z’s are an elusive audience that are least receptive to video and audio advertising. Looking across generations, Millennials are receptive to both audio and video ads, although video reported a four percent higher receptivity to video, whereas GenX are 32 percent more receptive to digital audio than digital video advertising.
Digital Video viewers are highly receptive to ads when spending time with family and dramatically less so when pursuing interests and hobbies. Audio listeners, however, stayed relatively situation-agnostic with receptivity levels remaining fairly even across different scenarios.
Gen Z, Millennial and Gen X parents are dramatically more open to ads than their childless counterparts by dramatic margins. For instance, Millennial parents are 27 percent more receptive to video ads and 15 percent more receptive to audio ads than those without children.
Audio is audio when it comes to ad receptivity with content format (music, podcast, audiobook) does not matter, unlike Video where content length impacts ad receptivity. Video viewers were most receptive to mid-length content like TV shows.
“Digital audio and video provide significant opportunities for advertisers to target audiences at the most opportune moments, but it requires really understanding what they are feeling and thinking when consuming different media,” said Kara Manatt, SVP, Intelligence Solutions & Strategy, MAGNA Global.
“People are focused when listening to rock music, excited when watching action movies… their mood states vary wildly throughout the course of a day and so does their openness to receiving an ad. Brands that understand the mood behind the action are dramatically more likely to grab the attention of listeners and viewers.”