Monetizing hard news has always been a thorn in the side of general-news publishers. For some advertisers, appearing next to a negative news article that describes a nasty terrorist attack or leans too much to one side of a polarizing political debate, will always be a hard no. But Prisa Media has a plan to alleviate that advertiser reticence.
The Spanish and Portuguese-language media group, which owns national newspapers El País and Cinco Días, has created a tool that uses machine learning to create contextually relevant and brand-safe audience segments across approximately 200 million ad impressions El País generates every three weeks. The newspaper generates more impressions than this, but this is the amount the team has ring-fenced as suitable for the tool, according to Pedro Ventura, director of technology on data and monetization, at Prisa Media.
“News can’t be happy all the time; that’s the reality,” said Ventura. “We’re general news, so like the Guardian or Le Figaro, we cover a broad spectrum, and, of course, much of that is terrorist-related news or the Catalonia conflicts. Some advertisers just don’t want to be near that.”
But the way hard news is currently defined is too blunt, he added. For instance, a story about a lost child that’s been rescued shouldn’t necessarily be lumped in with hard news and not open to advertisers that want to appear alongside content that evokes a feeling of happiness or satisfaction with the reader. Currently, verticals in which many advertisers are particularly sensitive to any hard-news item, like telcos and luxury, simply don’t advertise across a large proportion of El País.
The publisher has a dashboard that isolates the articles — of the 15,000 articles published each month on El País — that are potentially risky for brands. To determine how people respond emotionally to its articles, the team worked with market research firm Cocktail Analysis to monitor the responses to articles supplied by 2,000 participants. That data is then fed into the algorithm. This method will be used for the next year. “Having that human element is important in order to train the algorithm,” added Ventura.
So far the publisher has created 32 audience emotions and created “happiness” segments among others, to put advertisers at ease. It will charge a higher fee for those that want this option on top of regular targeting. So far, it has pitched the tool to some major brands that have shown interest, and tests will start in the coming months, according to the publisher.
Publisher tools like this are welcomed by agencies. “A machine learning-driven approach presents an opportunity to capitalize on the power of context at scale while ensuring brand safety,” said Ryan Storar, svp and head of media activation at media agency Essence. “We’re keen to understand the effectiveness of solutions such as this at scale in terms of delivering brand outcomes.”
It’s not always a clear-cut response. Responses to political topics like Brexit or others like government corruption can have polarizing effects and are subjective. That’s why the team will track the responses of between 20 and 30 participants for each article. If responses are split they will create an average across the responses, and feed that back into the algorithm. If the responses are too polarized the team will dismiss any impressions around the article.
Such a large technology investment is a lot for a legacy media publisher to shoulder. The media group applied for additional funding from Google’s €150 million ($167 million) Digital News Initiative. Publishers can receive up to 1€ million ($1.1 million) for large projects, €300,000 ($335,000) for medium ones, and €50,000 ($56,000) for smaller prototype schemes via Google DNI. Prisa won the middle-tier category, which will fund 70% of the entire project, according to Ventura.
Prisa dedicated four developers to the project, and worked with 10 people externally spanning expertise including data engineering, data science and staff from Cocktail Analytics and tech consultancy Indra.
Although Prisa Media and El País generate subscriptions revenue, advertising represents the majority. Digital ad revenues accounted for 53% of total revenue in 2018, and grew 13% compared to the previous year — while print advertising dropped 11.5%, according to the company’s last financial statement. In 2018, Prisa Group, which spans several continents including Europe and Latin America, and includes radio stations, generated €228 million ($255 million) in the first nine months of 2019.
While general-news publishers have always wrestled with the challenge of monetizing hard news, the continued decline of print advertising, alongside the competition from Google and Facebook for digital display advertising revenue, makes finding alternative product solutions to appeal to cautious advertisers critical, added Ventura.
Publishers have experimented with different ways of addressing the challenge of monetizing hard news for years. Developing new lifestyle and entertainment or sports verticals is the well-trodden route, in order to open up inventory for advertisers that feel safer targeting ads around articles in those environments. More recently, a string of publishers, including the New York Times, ESPN and USA Today, has also rolled out ad products that they claim can match ads to people in certain moods. The BBC has also experimented with tracking emotional reactions to ads for years to prove the value of its branded content.
In time, Prisa hopes to develop English- and French-language versions of the tool, which it can then license to external publishers, potentially opening up a new revenue stream.