RealEstateAlerter - Using AI to find stories in property data

Learn how Team RealEstateAlerter of the 2024 JournalismAI Fellowship collaborated across time zones, overcame challenges in data access, and applied AI to strengthen investigative real estate journalism

[Credit for image: Anton Grabolle / Better Images of AI / AI Architecture / CC-BY 4.0]

A morning jog past Detroit's train station sparked an idea that would eventually bring together six journalists and technologists from across three continents. This seemingly simple moment would become the golden thread that brought together an innovative approach to real estate journalism. 

Despite different backgrounds and time zones, they had a shared challenge that would unite them for a months-long journey of collaboration, innovation, and friendship.

Anjanette Delgado, Managing Editor at Hearst, Amelie Sutton, Product Manager for News Automation and AI at Gannett, both, from the USA, Zahra Naghsh, Senior Data Scientist, and Patrick Dell, Senior Visuals Editor at The Globe and Mail in Canada, Hans Jørgen Roe, Editorial Developer and Andreas Fosse, User Data and AI Lead from Norway’s E24, formed the RealEstateAlerter team as part of the 2024 JournalismAI Fellowship Programme, which the Google News Initiative support.

Where It All Started

"It was like love at first sight," recalls Fosse, jokingly, describing the moment their team formed during the Fellowship's matchmaking process. What began as separate ideas about stock markets and data outliers quickly formed around a shared idea: to build a system that could flag potentially significant real estate deals by analysing patterns in the data. 

"As soon as we met as a group, we thought, ok, this is going to work," Delgado recalls. "And over time, that didn't change. It just got deeper and better," says Delgado.

Team RealEstateAlerter at the LSE during the 2024 JournalismAI Fellowship Meetup

The project had an investigative journalism approach, which Naghsh describes as "the sweetest part of AI because it sort of minimises the risks and maximises the benefits at the same time." 

"The ideas that we had, were close to each other to some extent, which is why we decided to come together in the first place," Naghsh explains. Adding, “We agreed on a common idea, but the core of everything from my point of view was sort of an investigative viewpoint into  AI." 

RealEstateAlerter grew from an observation that important real estate transactions often contain newsworthy stories, but journalists typically discover them through chance or time-consuming manual processes. 

Delgado explained the original idea for the tool stemmed from a real estate story at the Detroit Free Press. It all started when an editor, while out jogging, noticed how the sale of a train station had dramatically shifted the local housing market. "I saw how important it was to tell that story, but also how human it was," Delgado recalled.

Over time, she began thinking about whether a systematic approach could be used to find similar stories hidden in real estate data. "This is data and numbers; couldn't we apply some sort of technique to get it? " As the idea took shape, she saw the Fellowship as an opportunity to explore it further.

Why join the Fellowship

“It was a chance to learn from other technical people and journalists working with AI in their newsrooms," says Dell. "And to learn from my colleague Zahra, while seeing how different teams were tackling the challenges AI presents."

For Delgado, it was the opportunity to collaborate and test an idea she couldn't explore on her own.  "[The Fellowship] allowed me to take that idea forward and think, how might we do this? It turned out so much better than I could have done on my own, which I probably couldn't have and I wouldn't have thought of so many things we ended up doing”.

Delgado says additionally she wanted to work with smart people but later found they were not only smart but incredibly kind and generous. 

Fosse says he had been following JournalismAI for a while and saw the Fellowship as the perfect place to develop his ideas. "As a small newsroom in a niche, it was a great opportunity to find like-minded people and work on something bigger than what we could have done on our own," he says.

Beyond Technical Skills

While the technical achievement was significant, team members consistently highlighted the human connections as the most valuable outcome of the Fellowship.

"I had limited experience working with a big group like this," admits Naghsh, who described herself as initially nervous about networking. But as the Fellowship progressed, she felt more comfortable and through the relationships, says she gained a meaningful experience." Adding that the most valuable part of the Fellowship for her was not only learning technical skills but also helping her develop soft skills. “It felt like a work family,” she says.  

Sutton says her prompt engineering skills improved dramatically as she worked on front-end coding. “Everyone will probably echo this, but working with Zahra, I learned a lot, so much about the sort of pipeline that you were building. And I think that was probably the biggest takeaway, just working with such a stellar technical partner was life-changing,” says Sutton. 

Fosse says he gained insights into data analysis methodologies and implementing AI in newsroom workflows. “Through practical implementation, we worked on concrete tasks while also learning from discussions with other Fellowship groups,” says Fosse. 

The Fellowship created an environment for collaboration, says Fosse, allowing them to work together on AI-related tasks and helping with developing problem-solving skills through practical implementation.

Delgado says the Fellowship sparked a side project, a list of notable people that could be cross-referenced with buyer and seller information. “I started to think about AI not as a single project that only works for one thing, but as an interconnected network of projects where we could start to use this list of names in other ways as well.  I could see it being a real ecosystem more than I had thought coming in,” says Delgado. 

Challenges Across Borders

Building the tool wasn't without challenges. Early in the process, the team spent significant time on organisational structure and establishing effective communication across three time zones. They maintained regular meetings despite the time differences and relied heavily on communication channels such as Slack.

Data access proved particularly challenging. While Norway offered comprehensive real estate data, accessing similar information in North America was expensive and less accessible. In Canada, access was further limited by proprietary data sources connected to the real estate industry. This led to the decision to focus primarily on Norwegian data for the initial prototype.

Once they had the data, cleaning, and feature engineering consumed over half of their project time. "We had to interview editorial experts, reporters, and real estate reporters," Naghsh explains. "It was not just within our team. We needed output from other sources and different newsrooms."

The Breakthrough Moment

The team faced a moment of near-disappointment when their first attempts yielded too many false positives, the AI was flagging many transactions as newsworthy that simply weren't.

But persistence paid off. The team worked intensively to refine the system, improving the few-shot prompting examples and fine-tuning the prompts. "That had an immense effect," Fosse recalls. "It was quite cool to see what you could do just by refining the prompt."

Lessons for Cross-Border Collaboration

For news organisations considering similar international collaborations, the team advised:

“You get what you put in,” says Fosse. Explaining that it is important to create an environment that encourages everyone to participate.

Dell says their “discussions and inputs were equitable and led to some interesting discussions around solutions for some of the challenges we were facing”. 

Delgado emphasises the importance of establishing organisational structure early, while Sutton highlights how dividing work across time zones can become an advantage, with team members picking up where others left off.

Sutton says a shared agreement among the team, that they were all working toward something together, made the collaboration easier, especially working remotely.

Most importantly, they stress the human element. "Be human. Learn to know each other and ask how things are going," says Fosse. Delgado agrees: "We talked about vacations and kids and all sorts of things, it helped us get to know each other."

Lastly, “find a concrete objective," advises Fosse, noting that a clearly defined, achievable goal was important given their limited time resources. 

A Lasting Impact

"It's been impactful and something I will remember for the rest of my life," says Fosse. "It's impactful in so many ways,  on the human level, networking and meeting other people, gaining better project management skills, and seeing how to implement AI into the newsroom."

Naghsh echoed Fosse’s sentiments, "That human aspect of this whole collaboration and teamwork makes big things happen from a technical point of view and other achievements. This was one of the best things I've ever done – joining this group and working with these people."

The highlight of the Fellowship for the team. “London was fantastic. "It was a great bonding experience for us,” says Delgado 

At the end of the Fellowship, all the teams met up in London. And when the RealEstateAlerter team finally met up “it was just like old friends by that point,” says Delgado. 

The rest of the team shared the same sentiments.

The Future of RealEstateAlerter

The story doesn't end with the Fellowship. E24 in Norway has already begun implementing the tool, with plans to expand access beyond the small group of journalists currently testing it. "I think it's going to have quite a profound effect on our ability to gather newsworthy real estate transactions," says Fosse. 

The American team members plan to follow Norway's lead, potentially adapting the tool for US data later in the year. And despite being "kicked out of the Slack channel" when the Fellowship ended, Fosse says they’ve created a Discord server to continue the conversation on the future of RealEstateAlerter. 

Delgado says, “We'll find a way to stay in touch no matter what.  As we work through the next steps for the RealEstateAlerter, we'll always stay in touch just to ask what each other is doing.”

Watch Team RealEstateAlerter’s presentation at the 2024 JournalismAI Festival.