Blocking and flagging inappropriate content through AI-enabled contextual outcomes engines is the way of the future. This new method replaces old algorithms that simply rely on keywords, which alone do not protect a brand entirely.
A recent whitepaper by Factmata and 4D (a division of Silverbullet), Is Brand Safety Enough?, explores the many challenges that reside within the (m)adtech landscape, and the issues existing contextual technologies face in brand safety and the high risk to brands associated with their failure.
The paper found unsuitable content flagged by Factmata, but missed by existing brand safety vendors, equates to 0.71% of total spend. With global programmatic spend in 2020 reaching
US$126.5 billion, advertisers have spent up to $898m on content considered unsuitable or unsafe for brands. This is a lot of wasted budget.
The paper also reports on whether existing available brand safety filters are applied to campaigns or not. Factmata’s technology will consistently block 4% to 5% of the total budget being currently spent on content at risk of Hate Speech, Propaganda, Sexism and Racist Content, which is unsuitable to brands and advertisers.
How much of your budget are you wasting?
Factmata is a London-based AI-company that helps brands, publishers, and platforms focus on nuanced brand safe environments by giving them a deep understanding of the quality, safety and credibility of any piece of content on the web.
Unlike traditional solutions which rely on keyword algorithms, Factmata’s machine learning tool accurately scores online content against eight different signals, giving each page a rating to determine how much of that content matches each score. These signals include racism, hyper-partisanship, fake news, sexism, personal insult, threatening language, toxicity and obscenity.
Silverbullet and Factmata set up a campaign on The Trade Desk to measure existing brand safety tools. Ad units in both line items contained a privacy-compliant JavaScript tag that allowed for the full path URL of all impressions to be collected and scored by Factmata.
Following the activation, 4D’s Data Science team and the Factmata team analysed the impression-winning URLs, scoring these pages against Factmata’s brand safety signals, to determine a Factmata Trust Score per URL.
Leading brand safety tools consistently missed unsafe content
According to the whitepaper, there were multiple examples of unsafe content, which was blocked by Factmata, but not identified by leading brand safety tools.
Below are just a few examples.
According to the whitepaper, Partisan content (0.88) and Hate Speech (0.68) both score very highly on each signal. The content – although in itself not advocating anything serious – presents a one-sided viewpoint and can thus be described as being partisan.
The article discusses different slogans, most of which revolve around Hate Speech, hence its relatively high score. Brands and advertisers would want to avoid this content as it can convey association between the opinions given and the brand advertising. However, the measured tools did not block this content.
In the future classifying various news pieces as low or medium risk, according to the Global Alliance for Responsible Media (GARM) risk categories, will allow brands to have the appropriate control over where their advertising appears.
As you can see, because the vast majority of these traditional brand safety solutions were built at the dawn of programmatic advertising, many have not kept up to speed with the nuances in how harmful online content is framed, worded, and produced. Dated algorithms rely on keywords, which alone do not protect a brand entirely.
About the research
4D’s Media Activation team ran a series of A/B tests to compare the effectiveness of traditional brand safety solutions at identifying and blocking complex suitability signals. A test campaign was set up on The Trade Desk, with a significant spend budgeted for two line items: brand safety applied and brand safety not applied. Ad units in both line items contained a privacy-compliant JavaScript tag that allowed for the full path URL of all impressions to be collected and scored by Factmata. All other targeting, bidding and media variables were identical. The campaigns ran for three full days in order to gather enough impression-level data to ensure the results were statistically significant.
About Factmata
Factmata is a contextual artificial intelligence company that builds technology to understand the nuances of online content, far beyond sentiment analysis. It has built natural language processing algorithms to detect multiple categories of content within articles and short form comments or posts, including propagandist tone, sexist language, hate speech and more. It provides solutions for brands to avoid online risks and reputation damage, advertisers to perform effective, safe ad placements, and many more. Founded in 2017 out of the UCL Machine Reading Lab, Factmata’s team consists of PhDs and data scientists across the globe, and is backed by Mark Cuban Companies, Biz Stone (co-founder, Twitter), Craig Newmark (founder, Craigslist) and Mark Pincus (founder, Zynga).
About 4D:
4D is the leading contextual outcomes engine that enables clients to step into the post-cookie, first-party data future with confidence. We bring together the most advanced machine learning and AI technologies to help you reach your customers at the right place, right time, and in the right moment. 4D is a subdivision of Silverbullet.
About Silverbullet
Silverbullet is the leading marketing transformation company that helps the world’s biggest brands to improve business outcomes. We reduce the friction across digital marketing, empowering better, faster and smarter marketing decisions. Founded in 2016, the company is headquartered in London, with a team of consultants, product experts, data scientists, and marketing engineers in Milan, Munich, Melbourne and New York.
Comments are closed.