AppliedXL and Associated Press team up to use AI to sift through federal regulations

In an era when local newsrooms are stretched thinner than ever, keeping up with the constant deluge of government regulations and updates can feel impossible. As federal regulations only continue to grow in complexity and volume, journalists face the challenge of not only sifting through the data but in discerning what matters most for their communities.

That’s where AI comes in.

Last month, the Associated Press and computational journalism startup AppliedXL announced AP Local Lede, an AI-powered tool designed to help journalists sift through large datasets and identify actionable local news tips. The tool analyzes data from federal agencies to extract insights that highlight how federal regulations and government actions impact specific communities and businesses across all 50 states.

“What we are building with AppliedXL is a real-time information company that leverages computational journalism techniques,” AppliedXL CEO Francesco Marconi tells Fast Company. “Combining editorial standards with AI, to surface novel signals that otherwise with human workflows you would miss.”

Local Lede is powered by algorithms integrated with input from AP journalists that analyze data from trusted public sources, provide context, and perform programmatic data validation and self-checks.

The pilot program has started with five AP member news organizations, with plans to expand to one still-undecided state in the coming months.

At its core, this technology is based on AppliedXL’s proprietary Live Ledger, which extracts events and risk signals from vetted public data sources. Marconi says this is a way of indexing and interpreting different sources of data, essentially organizing the world of regulatory filings and company disclosures through a journalistic lens.

Datasets are tailored for the sector. Local Lede is focused on regulations, and includes data from over 430 federal agencies, including the Federal Register, the official daily journal for rules and notices of all federal agencies and organizations.

“[The Federal Register] is a lot of noise, like every little incremental detail gets recorded. Reporters just don’t have time to scroll through all that,” says the AP’s Director of AI products and services. “When they have, they’ve found some real nuggets. There’s real news value in there. It’s just hidden amidst all this noise.”

Say, for example, a recent update from the Federal Register about a new Supplemental Nutrition Assistance Program program introducing advanced technology to streamline food stamp benefits, starting in select states. AP Local Lede would surface this tip, giving those impacted communities the opportunity to report on how this new program would affect the specific communities in their audience, according to Thibodeaux.

Marconi says AppliedXL’s “secret sauce” is their internal News Automation Engine, which allows the company to define a specific point of view in plain English to the language model and set analytical and programmatic benchmarks. The AP gave AppliedXL journalism guidelines and best practices, as well as a list of AP journalistic principles.

AI agents act like analysts with a journalistic lens, sifting through raw data, recognizing outliers, and assessing the significance of changes. “It’s sort of a news detection engine that is looking for very particular attributes, and it’s looking for patterns and outliers,” Marconi says. “Based on different criteria that are defined by journalists, we determine whether or not we should consider that a candidate for a story.”

For instance, in the biotech sector, a significant change in the number of people a pharmaceutical company is treating could be flagged by the AI agent as newsworthy. If the number suddenly increases, it might indicate that the company is struggling to achieve the statistical significance they hoped for in a clinical trial, a potential early sign that a drug is not performing as expected. The AI agent would then compare the situation with historical data on similar drugs at similar stages, determining whether the change is newsworthy.

Information deemed newsworthy is generated into a news tip – a short brief with newsworthy facts, relevant context, and the relevant communities or businesses that would be impacted.

The last step is fact-validation, where the AI generated report is compared to the original data source. If the facts align, the tip is distributed to the AP Newsroom for journalists to view.

The AP’s Thibodeaux said they are continuing to iteratively refine which tips are worth covering.

“There are still too many tips in there that are going to be a no, and we need to hone that down,” Thibodeaux says.

The team is also working on improving the quality of the context and headlines generated, as well as making sure that contact information is always included in the tip when available, according to Thibodeaux.

This partnership comes at a critical time for local news. The number of local news outlets continues to decline, and local newsrooms are struggling with diminished resources and staff. Interested in AI, but lacking the resources or time to experiment with it, Local Lede offers local newsrooms a lifeline—an AI powered tool that can monitor in real-time all federal regulations to generate timely leads for readers.

AppliedXL has to date raised $6 million in seed funding since its founding in 2020. Marconi says the 12-person team, based in New York, is in a “strong growth phase, marked by substantial revenue increases driven by adoption of our platform and APIs.”

In addition to the collaboration with the AP for U.S. regulatory risk and compliance, AppliedXL inked a partnership with Bloomberg in May 2024 to deliver real-time tips in biopharma and drug development. AppliedXL is currently the largest provider of algorithmically generated clinical trial and drug development real-time information.

AppliedXL is starting with “highly regulated” sectors, which tend to have the most substantial digital footprint, according to Marconi.

​​“It’s about being able to reliably monitor spaces like healthcare or regulation or energy,” he says. “Rather than reacting to breaking events, being able to have the early knowledge to be prepared for that.”