The leap from AI experimentation pockets to broader adoption delivering tangible business benefit is one of the greatest challenges businesses now face – publishers, especially. And it’s one that Mediahuis, the multi-platform multi-national media group, is tackling with an ambitious and consolidated AI strategy.
In 2023, the company drew up a strategic plan to shift from a 70% print margin to a 70% digital margin within seven years, with AI and automation central to achieving this.
The thinking was (and still is) that AI can help free staff from repetitive tasks – allowing more focus on journalism and strategic innovation, with automation also playing a key role in print operations and distribution – helping to cut costs and enabling reinvestment in digital products.
An AI team was established. And experimentation began.
Then, in June 2025, its first head of AI strategy – Ana Jakimovska, a former director of product management for Guardian News & Media, whose CV also includes consumer-focused digital product roles at Sky, BBC Worldwide and Which? – came on board.
And over the months that followed, efforts stepped up further.
“When I joined, it was a strategic role without a delivery side,” she recalls, speaking one year into her role.
“It was looking at the organisation, the industry, the challenges and opportunities ahead to formulate a consolidated AI approach – initially, focused on a three-to-five-year horizon – that would allow us to benefit from AI disruption in the longer run. This has since evolved to look at mid-term and long-term opportunities and risks.”
Mediahuis had already built a small but growing list of AI use cases and established an internal network to support internal AI queries and share AI experience of different tools and applications and best practice.
Next, it needed to identify enablers to set itself on the right path – at a time when while AI and AI adoption was fast-growing, its future shape and associated challenges and opportunities remained unclear (and still are).
“The way we started thinking about it was portfolio management, given that uncertainty is so high,” Jakimovska explains. “And we devised a few routes forward based on where the highest probability for success for publishers looked like it would be given the market – essentially to be ready for those should they arise.”
One immediate priority was safeguarding value by protecting Mediahuis’ journalism from being used by AI without value exchange.
In the short term, this has involved implementing technical protections. Over the longer term, it means working at a business and industry level to secure more responsible use of publisher content by AI.
“We are now at the point where, when we test, almost nothing gets through in terms of malicious bot activity,” she says.
“Moving forward, it’s about taking publishers beyond needing to spend lots of money on their own protections to a position where all players in the industry, especially AI companies, have bought into more responsible use and fair value exchange.”
Upscaling experimentation
Another priority was to upscale experimentation.
The first step was raising the 4,000-strong workforce’s knowledge and understanding of AI to the highest level.
“Over the past year, we’ve done three phases of this and are now moving into the fourth – essentially, automating workflows and gathering use cases from around the company of cases where employees believe AI can help and free themselves up to do higher level tasks,” Jakimovska continues.
“We’ve been very aggressive with this – it is mandatory and continuous – and, in this respect, a little ahead of the game.”
So far, 177 use cases in areas which AI has potential to optimise – such as audience engagement, monetisation, sales conversion and upselling as well as personalisation of B2B and B2C propositions – have been gathered, clustered, prioritised based on impact and implementation of them is now underway.
“Experimentation is brilliant but equally, it’s also difficult to monitor in terms of return, so one part of our AI literacy and adoption programme is about what you need to gauge impact – what value an AI application has delivered – and what safe data looks like,” she adds.
The second step was upscaling support to further empower people to experiment.
An emphasis was placed on recommending tried and tested tools, for example, and encouraging in-house entrepreneurial efforts towards applications that don’t just benefit one individual or team’s day-to-day requirements but can be scaled business-wide.
“We want people to experiment, but a tech stack can become quite fragmented. One way to address this is to recommend tools that have satisfied pretty much all the use cases we have seen,” Jakimovska explains. Meanwhile, those developing more complex use cases work directly with one of Mediahuis’ technical teams.
A big challenge many businesses face when scaling AI experiments is lack of understanding of the cost of doing so – notably, the token cost (tokens being the basic unit of text an AI LLM processes and charges against).
This can be improved by ensuring those involved in any project are supported by an AI expert to sense-check the cost-benefit so all understand the true business value, she says.
But also important is to bridge the inevitable lag that occurs between upscaling investment to roll out an AI application at scale and that upscaled AI application delivering business value and tangible ROI. To cover this gap, parallel cuts within a business will also be needed.
“There are some established methods and techniques for cost management within closed foundational models (large, general-purpose AI models trained on vast amounts of data, such as Claude or Gemini), and the companies providing these are aware and mindful of this problem with a promise for a solution,” Jakimovska continues.
“An emerging area is using open-source models – when their quality maps to the use case – as a cost management strategy. For now, our approach is being cost-aware from the onset of the agentic use, using established techniques to minimise (cost) and focusing on the use cases with the highest return.”
Special projects
Another important component of Mediahuis’ AI strategy has been identifying “big bold bets”, which has led to a number of special projects now also underway.
One is exploring how and at what level AI can be used to automate ‘first line news’ with AI agents deployed to select and prepare stories from a database of key news sources – such as parliaments, wire agencies, companies and think tanks – according to the different wants and needs of different Mediahuis brands’ audiences, and track response.
Announced earlier this year, the idea is that automating the routine, foundational ‘first line’ reporting of events in this way will free up more time and resources to spend on the high quality deeply reported ‘signature journalism’ audiences want, value and trust.
The approach is to use AI within content (as opposed to workflow) only for first-line content and always with strict human oversight and ethical safeguards.
Scaling AI is best done within a clear, consolidated and strategic framework, Jakimovska believes. And a dedicated head of AI strategy – a function which in many organisations is split between multiple operational roles, she says – brings greater focus and clarity to AI planning.
Not putting all our eggs in one basket…
It has also allowed Mediahuis to start considering issues she says she sees few others directly addressing – such as technical sovereignty.
“At the moment, we are all using US technology. And while those companies are brilliant and their products are fantastic, the question remains what’s the risk and vulnerability of having all your tech supplied by just three or four people?” she explains.
“There’s also the question of data sovereignty – how best to protect our data and our subscribers’ data. And, as a European business, the question of what relationships we have with European suppliers to support the European economy.”
Mediahuis’ consolidated AI framework has also enabled the business to be more active at an industry level.
In May, for example, it became a founding member of the SPUR Coalition which was set up to help set global AI standards for journalism with clear rules and technical standards for how AI platforms use journalistic content – focusing on transparency, licensing, and fair compensation.
“As an industry, we need to come together and work collaboratively to define standards that publishers – and AI companies – will be comfortable with. And we need to engage with the right companies, bodies, policies to ensure the right standards are implemented at pace before it’s too late,” Jakimovska believes.
Just how AI will evolve further from here and what this will mean for publishers is not certain, of course. Even so, the time to prepare for the possible scenarios now emerging is now.
“One possible scenario is to remain a destination focusing on ensuring your journalism is original and unique. The human part of your journalism – knocking on doors, investigating, connecting communities – that AI cannot replicate is your competitive edge. And you harness your strong relationships with your audience with great experience,” she says.
“Another is your content is consumed by agents. You are no longer a destination and your battle is for fair value exchange and brand recognition. And because your content is high quality and respected and your brand trusted, you find a way of working collaboratively with tech companies because everybody values and benefits from trustworthy information.
“Additionally, there are efforts in creating an emerging content marketplace in which the high information density that comes from journalists’ output is repackaged in different ways for different use cases. However, how this might work, for what products and at what prices is not yet clearly defined.”
Across any of these scenarios, publishers need to act now rather than wait for a tech company to define tomorrow’s content marketplace and how it will work. And they should do so by being a co-creator of the future value exchange, distribution and delivery of publisher content, Jakimovska adds.
Because while what tomorrow’s AI world will look like is unclear, tomorrow’s successful publisher is.
“A successful publishing business in the near future will be one that is creating revenue and thriving commercially,” she believes.
“It will have fully embraced AI while maintaining its trustworthiness and retaining its core values. And it will be deeply embedded within its community – having built a deep understanding of its audience, who they are and what they want and shifted its products to attract a younger generation.
“Publishing has been through a lot of change and transformation and problems in recent years. But honestly, I think that now and whatever lies ahead is an incredibly exciting time – a time with more opportunities to be found than actual risks.”
This article was first published in InPublishing magazine. If you would like to be added to the free mailing list to receive the magazine, please register here.
