With the endless expansion of the digital content universe and growth of video advertising, many companies are leveraging contextual targeting to place ads alongside relevant content, as opposed to targeting ads based on user data. Zefr bridges the gap between the vast amounts of content out there, and companies seeking more nuanced contextual advertising solutions.
Zefr’s technology platform uses brand preferences to source relevant video advertising opportunities from platforms like YouTube and Facebook. With deep insight into video context and how to maximize brand value, Zefr enables content targeting that is both precise and scalable based on each brand’s nuanced contextual preferences- a custom solution for every customer. Their positive contextual targeting solutions can amplify the right message, at the right time, and help companies meet their advertising campaign objectives.
As more brands started to scrutinize their content and turn to Zefr for contextual advertising solutions, the company needed a sustainable way to scale operations. Initially, Zefr started an internal crowdsourcing program to review and categorize vast amounts of video data, reviewing approximately 30,000 videos over two months.
Quickly, it became clear that this internal program was difficult to scale. It had limited quality control measures and not enough reviewers to confidently rely on the data labeling results. This led to more manual review work near the end of the process before a recommendation could be given to a customer.
Zefr started looking for outside help in improving the quality and increasing the output of their data insights so that they could process the increasing volume of data for their customers.
In searching for a solution that wasn’t overly-engineered, but was cost-effective and flexible to their evolving needs, Zefr turned to Appen in 2018. With Appen’s crowdsourcing solution, Zefr suddenly had access to the large pool of people ready to label data. More reviewers working more efficiently meant machine learning models could be trained quickly to output accurate video recommendations.
Initially, Zefr worked with Appen only on brand safety efforts (i.e., eliminating content that is universally bad for every client), before expanding the partnership to address the challenge of customer-specific, nuanced preferences.
Today, Zefr’s context DMP (Data Management Platform) lets brands provide their contextual preferences, which are then stored and amplified using Appen’s platform to fuel Zefr’s model for each client. It begins with a team of Zefr moderators, who create jobs on the Appen Platform using brand preferences in the context of what’s relevant in video. Moderators then source content, create test questions around that content, and distribute the questions and videos to reviewers. Machine learning models process reviewed data and output relevant and high-value videos for specific customer brands.
“What’s really key is the guaranteed throughput more than anything, because that means I can give customers a definitive deadline, which I could never do before we started working with Appen. And I can get back to them with quality and quantitative numbers.” – Jon Morra, Chief Data Scientist
When Zefr crowdsourced internally, their team reviewed 15,000 videos per month. With our help, Zefr has been able to process through about 100,000 videos each month, or 6.6x more data per month, allowing them to focus on their business and growth. The quality and quantity of the data produced have improved tremendously, to the point where the Zefr team feels secure in passing along solutions to customers without spending extra time re-reviewing work that machines and reviewers have already done.
Our crowd provides a consistent, predictable quantity of data output. For Zefr, that means being able to give customers more precise turnaround times with quantitative numbers, without sacrificing quality.