A small Norwegian newsroom punches above its weight with a data-driven, human-centred AI strategy

iTromsø is one of two local newsrooms serving Tromsø, a town located on a small island off the northern coast of Norway. Founded in 1898, the newspaper is part of Polaris Media, a media group with more than 70 publications across Norway and Sweden.

With an editorial staff of 25 and a total circulation of more than 10,000, iTromsø plays a significant role in the local community.

However, a few years ago the publisher realised that their larger, better-resourced local competitor was steadily out-reporting them, and that they were in danger of falling behind.

"We were losing the news fight. For every story we had one person on, they had four or five," said Lars Adrian Giske, iTromsø's Head of AI at our recent Paris AI Forum.

iTromsø's response was to develop a three-part, AI-driven strategy aimed at amplifying their impact and gaining a competitive edge:

Automating repetitive workusing machine learning to free journalists' time for reporting and writing.

Investing in data journalismto analyse publicly available datasets and "create our own news that are locally relevant," Giske said, instead of chasing the same news as others.

Doubling down on fact-based reportingand contributing to public debates with original, data-driven stories at a time when opinions risk overshadowing facts.

"We believe that facts matter, and that we can provide those facts and help anchor local discourse around facts," said Giske.

Speeding up local investigations with AI

The first outcome of the new strategy was Our City, a reporting project combining data from various sources, including tax records and property and car registries. The goal was to see "what this data tells us about the place we live in," Giske explained.

As part of the project, the team uncovered and told stories about a "massive hidden inequality from area to area in the city," Giske said.

This was followed by a similar initiative based on fishery data, which revealed fraud and illegalities in the locally important fishing industry.

While these projects helped iTromsø to produce a wealth of original, high impact journalism, they also required a great deal of manual labour. It was clear the newsroom needed a more streamlined solution for future projects.

In collaboration with IBM, the publisher created a new data platform focusing this time on urban development. The goal was to help journalists find potential stories in the digital municipal archives, which were very difficult to navigate.

"We quickly identified this as a major time sink that journalists were spending time on. They were spending two to three hours going through this archive instead of producing news," Giske said.

The result was DJINN (Data Journalism Interface for Newsgathering and Notifications) - a platform that extracts documents from municipal archives, summarises them, and ranks them according to their newsworthiness based on a scoring system developed with the input of journalists.

Using the new tool, journalists found relevant documents much faster: "Instead of spending two hours doing that work, they spend five minutes doing it and then start calling sources and working the story," Giske said.

Indeed, journalists jumped on the new tool as soon as it became available, and in just one week, the system helped produce six cover stories for iTromsø.

Examples of articles that iTromsø was able to report on thanks to its in-house DJINN platform.

"This is freeing up time to do the important work, the human work, which is going out and talking to people. That's how you make this kind of news relevant to our readers," Giske said.

"We started out missing out on stories and being beaten by our competition. Now we are winning the news battle, and we have way more good story ideas than we know what to do with."

A 'federated' model for scaling innovation

When the iTromsø team presented the new tool to owner Polaris Media, the media group was immediately keen to scale the initiative and make it accessible to its other local newsrooms.

The scaling process started in August 2023, and by the following February, the DJINN platform had been rolled out in 35 newsrooms within the media group. Most have seen an increase in the production of articles related to urban development, as well as significant increase in traffic.

Combining data stories with a human-centric angle has been key. As Giske said, for a reader "it's much easier to relate to a piece about someone who is impacted" by new regulations, for instance, than to "reading a summary of a document from the municipality, which is what all our journalists had time to do before."

Alongside scaling the DJINN tool, the publisher wanted to establish a framework for developing and scaling new tools. However, the fact that Polaris Media follows a "federated" model complicated matters, as "every newsroom has full editorial independence," meaning the group cannot impose a solution on individual newsrooms, Giske said.

"There was no structure for aligning different newsrooms. They were all doing their own thing," he said.

"That doesn't work once you need infrastructure, and you do need infrastructure for AI or the kind of data work we're doing. It's not feasible for a small local newsroom to be responsible for all of it."

The group established an approach where it has five regional AI labs, each with their own specialisation and each embedded in a local newsroom, iTromsø being one of them.

"That means that we have developers at hand, and they are doing R&D on AI in-house in the newsroom," Giske said. "They're working very closely with the journalists every day."

The group also has a centralised AI & Product Forum that consists of representatives from each of the five labs as well as the group's development editor.

Key to this framework is that "the actual work, strategy work and development, is being driven by the labs themselves, by the newsrooms. And then we have product teams taking what we see works and we want to scale," Giske said.

In other words, Polaris Media as the group sets direction and infrastructure but does not operate by mandate. Thus, the overall model combines "bottom-up creativity, central coordination and top-down alignment," he said.

Scaling DJINN into an investigative engine

Building on the success of the DJINN platform, the team's next priority is building a data platform for investigative journalism. This will enhance the tool by combining data from various sources and automating initial research using machine learning and generative AI systems. It will also send journalists notifications and include an agentic layer for deep research and data visualisations.

While this may sound very technical, iTromsø's strategy puts people at its core. For instance, Giske does not believe that automated systems can produce credible journalistic output. Human journalists are still essential to the news production process.

"Journalism is a human-driven mission for transparency, access to information, and telling stories that matter. No agent system can do that at this time. They may produce things that look like journalism, but it's not journalism," he said.

And although many of iTromsø's stories may start from a dataset or a digital document, they always aim to put impacted people at the centre of their coverage. When done successfully, this engages readers on a deep level.

"How do we build trust amongst readers?" Giske asked. His answer: through stories in which audiences can see themselves, and which empower them to act and influence their local community.

Teemu Henriksson

Research Editor

[email protected]