While early uses of generative AI and ChatGPT have centred around editorial, news media companies are looking at how it can improve workflows and change processes. 

During this week’s Webinar, Insights into using GenAI,  Generative AI Initiative Lead Sonali Verma talked with Rob Aherne, director of the Imagine programme at Bauer Media Group, and João Matos, engineering and data science manager at Grupo RBS, about what they have learned so far on their respective AI journeys.

How AI can improve print processes

Bauer Media Group is Europe’s largest publisher of consumer magazines and the biggest commercial radio broadcaster. Although print remains a significant part of its business, the company has developed a digital strategy and began exploring GenAI to enhance efficiency in the magazine production process. 

“We’ve been looking at AI for a couple of years, and ChatGPT’s [emergence as the] breakthrough interface was the impetus for us to accelerate in this area and put some more structure around our activity,” Aherne said. 

Bauer formed working groups to define its approach, guidelines, and policy for AI usage, and conducted company-wide hackathons and proofs of concept to understand the potential of AI. 

“[Our] approach has been quite cautious, but it’s been anchored in trying to define the value of tools like ChatGPT in our business,” Aherne said.

At Bauer, that meant answering these questions:

• Where can AI tools be used in the workflow?
• Was it possible to quantify the benefits as well as the cost?
• What are the technology’s limitations?
• Would off-the-shelf tools do the job or would it need to be built?
• What training or onboarding would be required?

Regarding tools, Aherne said ChatGPT provided a logical starting point thanks to its accessibility and potential applications to support the editorial workflow. Aherne noted that while ChatGPT offers value to its editorial teams, especially writers, its usage varies across the organisation.

The most valuable use for ChatGPT, he said, is rewriting and repurposing content, as it saves time compared to manual tasks.

“We try to share and reuse content as much as we possibly can, whether that’s between brands or from printed digital or vice versa or sometimes across countries,” Aherne said. “ChatGPT is a real time-saver here.”

It also can be used to change the tone and style of an article to create different versions, which the company found valuable: “[We liked] the concept of being able to send one journalist to something like a red carpet event and then use AI to create new versions of their stories and their content to go out to multiple different brands really quickly.” 

However, Aherne emphasised that the output from GPT is never perfect and always requires human intervention.

If you build it…

To encourage greater usage across the organisation, Bauer developed its own interface that makes it easier for editorial teams to interact with the AI tools.

BauerGPT is currently in beta and used only in Poland. It offers users access to four tools: ChatGPT, a copy-editing tool, an image generator, and a transcription tool. Ahern stressed that no one should be manually transcribing audio in 2024 when effective tools are available.

Aherne also mentioned its challenges with getting AI to convert text-to-voice. Bauer is experimenting with creating AI-powered podcasts from archived print articles of its true crime magazine. The workflow involves extracting the article text, running it through ChatGPT to create a suitable script, and then using a text-to-speech AI application to transform the script into voice.

Next, Bauer is building BauerGPTs for specific editorial tasks, including one that writes tabloid stories in a specific tone based on information input by users. It creates an article draft written from the journalist’s perspective and in the requested tone.

“We don’t expect it to be perfect, but it still allows journalists to get the basis of the story together faster than doing it without assistance,” Aherne said. “So while they will have to alter and iterate it further and obviously check it when it comes to speed to market, we think this is hard to beat for certain types of stories.”

Grupo RBS uses GenAI for personalisation

Until 2018, Grupo RBS was focused on print B2C business, using data-driven strategies with statistical support for campaign optimisation and churn propensity modelling. Over the past five years, Matos explained, the focus shifted to digital B2C revenue, with machine learning models used for subscription and churn propensity modelling, and subscriber and lead engagement clustering.

The company recently began working on personalised news recommendation models and automated content production using GenAI.

“In the middle of last year, we put different areas together to think about internal opportunities of using Artificial Intelligence in the company, not restricted to generative AI,” Matos said. “Of course, at that moment we were in the hype of ChatGPT, so we were talking about generative AI.”

To explore internal opportunities for AI use, the company organised a hackathon focusing on content conversion, engagement, and customer service. The competition’s goals were to increase audience, optimise newsroom productivity, increase prospects and subscribers’ engagement, improve subscription conversion rates, and enhance customer service efficiency.

The hackathon resulted in three GenAI solutions:

• Base GreNal, which creates articles about youth soccer teams.
• Wingman, an AI “copilot” for the newsroom.
• GZH Escuta, which uses GenAI to interact with audience inputs across various channels.

Although these were initially “just ideas,” the company looked at their potential and began a digital transformation project in the second half of 2023 to increase its digital business over the next two years. The project includes 11 strategic initiatives covering B2C, B2B, and newsroom value streams, with significant use of data and AI.

Two strategic initiatives from the hackathon — Wingman and GZH Escuta — were incorporated into the digital transformation project. It launched a pilot project using GenAI for content production in December 2023, focusing on weather forecast news from structured data provided by an API.

The generated articles were initially reviewed by a journalist, but production and publication became 100% automated after three months. The project has been successful, with 15% of the audience in the climate section coming through the AI-generated news. Matos said a disclaimer at the bottom of the page tells users that the text was produced using a GenAI tool.

However, Matos said using GenAI for content production has raised SEO issues, as the increased volume of weather forecast articles can negatively impact the ranking of other articles. He said the tech team is working to resolve this issue.

Matos shared testimonials showing the project was well-received by the editorial team. They praised its ability to provide relevant content with agility and quality, and it puts the company at the forefront of AI use. Next, he said, “The plan is for this work to be replicated for our horoscope articles as well.”

Looking ahead, Grupo RBS plans to accelerate its digital business by growing consumption and maximising margins through data and AI. This includes content qualification, content personalisation, performance optimisation, and price value optimisation, with the first three using GenAI and the last using traditional AI models.

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