The Centre for Investigative Journalism of Serbia (CINS) faced a challenge many investigative outlets encounter: how to verify government claims backed by vast amounts of data when traditional manual processing simply isn't feasible. Their solution – a custom AI-powered tool – represents their first major step into artificial intelligence.

The problem: Debunking claims with mountains of data

CINS observed a surge of pro-government news coverage claiming dramatic reductions in healthcare waiting lists for major operations. The 11-person team, with seven working in the newsroom, suspected these claims were misleading.

"We thought that this cannot be true because it’s seemingly impossible for them to get thousands of people off the lists in such a short period of time,” explains CINS Director Milica Šarić. "So we started wondering, how can we debunk this?"

Their traditional workflow – requesting thousands of documents from Serbian institutions, scanning them, performing Optical Character Recognition (OCR), and manually processing everything – was becoming unsustainable. "We have done everything manually. We have a long history of collecting huge amounts of data,” Šarić notes. The JournalismAI Innovation Challenge, offered a timely opportunity to pilot a different approach.

Building the solution: When off-the-shelf AI isn’t enough

Initially, CINS attempted to use ChatGPT with scraped data, but quickly discovered its limitations. "It couldn’t answer all our needs. It couldn’t collect and analyse all the data that we needed,” the team found.

This led them to develop a custom solution with an external AI specialist who understood media requirements. "We decided to make something on our own – our AI specialist building our own tool that actually can bring in many Excel files, process them through SQL database imports and queries, and then provide responses like GPT when you put in a prompt."

The resulting tool functions like a specialised ChatGPT that can process multiple Excel files and respond to data queries.

The communication barrier

One of the most significant obstacles to building the tool wasn't technical but human. CINS discovered that the gap between journalists and developers poses a fundamental challenge for media organisations, particularly in the Balkans.

"That's a huge problem... finding appropriate specialists and tech people who will be able to understand our language, understand what we need,” Šarić explains. This affects the entire region: "Other media organisations that I know in Serbia encounter the same issue."

Their solution was pragmatic: working with their AI specialist to develop a shared communication approach that would make him more attuned to journalistic sensibility. "We worked with our AI specialist on the way that he needs to talk to us. We now have a new language between us."

Obstruction by design

According to CINS, Serbian authorities deliberately obstruct data access. "Sometimes institutions print everything and then scan it back in a way that obscures the text. They’re trying in every possible way to make it hard for us to use digital tools." The team also notes that the Freedom of Information Act has become increasingly unreliable.

These institutional barriers add another layer of complexity to AI implementation, as even the most sophisticated tools cannot process deliberately corrupted or withheld data. The team's custom solution can only be effective when institutions actually provide usable information or when such data is available online.

What they've learnt

The project revealed that AI integration requires organisational transformation. "We don't have a skilled enough organisation. So at some point we should bring on board more technical people, more digitally skilled individuals. But we also need to support and upskill our current staff as part of this journey,,” Šarić explains.

The editor-in-chief is now pursuing a master's degree in digital innovation management in the UK, planning to drive the transition to a modern digitally-oriented newsroom.

The opportunities: Enhanced efficiency and digital transformation

CINS views AI as a tool for efficiency rather than replacement. The solution processes data faster than manual methods and promises cost reductions by eliminating the need for additional staff for routine processing.

"This is an opportunity for us to produce more in a faster and more digital way instead of doing everything manually but significantly slower,” Šarić notes. However, they maintain editorial oversight and recognise that successful implementation requires sustained technical support.

Regional potential

The tool has universal applications for data journalism. "This tool can serve anybody who works with large sets of data. Anyone engaged with data journalism or investigative journalism can benefit from it."

The team is enthusiastic about knowledge sharing through masterclasses and conferences, viewing their experience as a foundation for broader regional AI adoption in journalism.

Lessons for newsrooms

• Bridge the communication gap: Recognise that the biggest hurdle is often the communication gap between journalists and technical specialists. Also, aim to bridge the gap between journalists and the AI solutions themselves, reducing the reliance on expensive technical intermediaries.

• Build custom tools for specific needs: Consider building custom tools instead of simply adopting existing solutions. This approach helps ensure the technology precisely addresses journalism's unique requirements.

For newsrooms considering similar projects, three factors proved critical: finding technical specialists who understand media, preparing for organisational change beyond tool adoption, and treating AI as a way to enhance rather than replace human judgment.