A news media company’s “editorial process” covers the distribution of content creation and acquisition to the audience. The audience is an important part of this process, as they give valuable feedback on the content.

Artificial Intelligence (AI) can help publishers cut costs, boost the user engagement with their portal, and help them thrive in the digital publishing ecosystem.

Yet when it comes to launching an automation process, Piotr Turek, head of data products at Ringier Axel Springer, told INMA members on Wednesday that process should not start with content creation.

“I believe it is a trap, and it’s best to avoid it, at least at the beginning. Instead, we very much prefer to focus on introducing automation into our content distribution processes. We consider that they have much more potential.”

On Wednesday, Turek led an INMA members-only Webinar about the automation and personalisation journey at Onet, the largest online media portal in Poland. Turek shared how to achieve goals with the automation of the editorial process, robots, and algorithms, and how new and innovative ways of using the data can create value.

Where to focus automation efforts and why

There are two reasons to focus on introducing automation into our content distribution processes, Turek said. 

1. High-quality content is king, and your content creators are one of your biggest assets. “Why should we try to automate them away if they are our biggest asset?”

2. Even if one wanted to automate content, there is a problem of hype when it comes to AI. “Even though our AI machine learning capabilities are growing very fast, the expectations towards them are growing even faster,” Turek said. “The expectations greatly exceed their true capabilities. They do no common sense reasoning whatsoever.” This makes it very difficult to create quality content through automation.

Therefore, Onet focused on automation — rather than creation — of content distribution.

“When thinking about the modern media landscape and the diversity of the audience that we are faced with ... they all expect heavily personalised content,” Turek said. “Because we live in this hyper-connected world of social media and Big Tech companies.”

In fact, Turek added that the big social and tech platforms are a publisher’s true competition, not other publishers. Those large companies have many advantages over most news media publishers:

• High levels of personalisation.
• They are winning the battle for user engagement.
• As a result, they consume the vast majority of the advertising pie.

“We concluded that we, as a professional media organisation, must respond to this challenge or we will face a slow death,” Turek said. He encouraged INMA members to read Ringier Axel Springer’s blog post on this topic.

Because much of the reaction towards AI and automation is that of fear (replacing humans and jobs), the Axel Springer team decided to put machines in the service of humans, rather than automating human jobs away.

A perfect place for this to occur was in managing content and ad placements on the home page of the Onet Web site.

“Before introducing personalisation, we used to have a dedicated person managing each of these widgets separately,” Turek said. This person managed the widget they were in charge of manually.

However, automating these widgets meant they could be adjusted faster, while following the dynamics of the various diverse users in real time.

“What we envisioned was an evolution of the role of an editor from a placement micro-manager to a much more interesting role to what amounts to a creative quality guard, focused on content itself,” Turek said. “Instead of managing the placements, they now have time to make sure the content is of sufficient quality.”

These editors control the editorial integrity, while the content they select is thrown into what Turek called the “magic cauldron” of the automation process for placement on the Web site.

“To make this work, you need a well-designed and well-functioning personalisation engine,” he said.

How to successfully personalise a news portal

Turek said that in Onet’s experience, the team learned there are several main pillars of personalisation in the news industry.

• Trust versus control. Ensuring there is an easy to use, fail-safe manual override to support editorial policy. The newsroom must be able to observe and react to article performance. Personalisation must support existing practices.
• Made to measure for the news industry. Publishing is very time sensitive, and publishers want to be able to be the first in breaking news. There is also the need to harness trends, as well as burst the “filter bubbles” that keep users trapped in a tightly controlled information sphere.
• Diverse business needs. Every widget can be a completely separate deployment and may want to be optimised for different things. The set of KPIs can be broad, even for just a single Web page.
• Costs at scale. A single version of a Web page served to everyone results in effective analytics and low cost. However, personalisation increases those costs and complicates analytics.
• Driving content creation. Content creators should be constantly informed through audience feedback. This is much more complicated, with far more information, in today’s age of personalisation and provides more data than the human brain can deal with.

The question then becomes: How can a publisher build a system that satisfies all these constraints? The answer to this will be the focus of another upcoming Webinar, Turek said, adding that there could be different answers for different organisations. He also shared another blog post that outlines how Ringier Axel Springer does it.

One thing Onet did was to reimagine analytics, to make them more straightforward. The process became:

• Descriptive: Say what happened.
• Diagnostic: Explain why.
• Predictive: Predict outcomes.
• Prescriptive: Suggest sensible actions.

“Instead of adding more and more reports, why not automate the process of going through those reports to draw conclusions?” Turek said. “And maybe as a result, solve the personalisation/analytics dilemma. That’s an area where we are investing heavily.”

This means editors can look to the system to easily find out what topics they should consider covering next to the widest, most engaged audience.

The bottom line for Onet, Turek said, was to concentrate automation first on content distribution, then on analytics.

A glimpse of Onet’s results and lessons learned

One of the biggest successes from Onet’s automation and personalisation journey has been an extreme improvement in content diversity. Today, three times more content reaches the users, and journalists are able to create more content.

“This all translated into very positive impact on user engagement,” Turek said. “We observed significant uplift in click-through ratios. We see that our users are more active on average. And they stay more loyal. So basically, the promise of personalisation was fulfilled in our case.”

In the end, the journey has proven well worth it. Onet has achieved:

• Extreme improvement in content diversity.
• Positive impact on user engagement.
• Business transformation that is being adopted by other publishers.

“If there is one thing that I would like all of you to remember from this talk is this: The editorial process automation should be all about employing machines to take over mundane, data-heavy tasks so that we humans can focus on unleashing the power of our creativity.”

Banner image courtesy of Markus Spiske and Adrianna Calvo from Pexels.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .