How Radar AI Transforms One Dataset into 300 Local Stories

PA Media's Product Development and Operations Director Martin Ashplant, Data Journalist Clara Margotin and Editor Lesley-Anne Kelly (right) explaining how Radar AI works. Not in the picture: Editor-in-Chief Jack Lefley

The British news agency Radar AI has built technology that amplifies output in ways that serve local communities while addressing the fundamental economics of regional news coverage. I visited Radar AI last month at their HQ in London as a co-organizer of the Chefrunde Study Tour. Our meeting revealed a unique company with a crucial role in the local news space.

Radar published 630 separate projects last year, generating over 38 million words of content. Editor Lesley-Anne Kelly calls this potentially making them "the most productive journalists in the country." Yet "every word we publish was written and edited by human journalists. Everything is checked by human journalists," Kelly emphasized.

Radar uses natural language generation (NLG) technology as a replication tool. One reporter writes one story, and the system adapts it across different regions using the same dataset. Where a traditional news operation might produce three or four versions of a story, Radar can scale that to hundreds. This distributed operation of just a few people spread across the UK reaches hundreds of local publishers.

Behind Radar's output lies the proprietary NLG system. Data journalist Clara Margotin walked us through how the platform transforms raw government data into localized stories.

The process begins with structured datasets from sources like the UK’s ONS (Office for National Statistics) or the NHS (National Health Service). "We take one data set that our national statistics provider puts out and produce 300 stories for local media using that data," Kelly explained.

The system uses conditional scripting logic: if Manchester's crime statistics have risen, it uses Script A; if they've fallen, it uses Script B. The system integrates area-specific information including demographics, comparisons to neighboring regions, and relevant local context. Built-in error handling manages missing data, confidential information, and statistical outliers that might skew the narrative.

When the ONS releases data, Radar can produce 300 unique stories for local media outlets from that single dataset. Each story is tailored to regional specifics, but the journalistic heavy lifting happens once.

Local newsrooms often lack the capacity for deep data analysis, especially when that analysis needs to be repeated across multiple geographic areas with similar datasets. Radar transforms hundreds of individual efforts into a single, scalable process.

Radar's content categories reveal the breadth of their coverage: fully automated daily stories about food hygiene ratings, weekly road closure updates (their most popular content with local media), and monthly housing price and NHS waiting time reports. Projects covering crime statistics, health data, and education metrics require human journalist input for context and quotes, with a typical two-day turnaround time.

The technology processes the underlying data to create genuinely localized stories. When the data shows unemployment rising in Manchester but falling in Bristol, the system generates appropriately different stories for each market, complete with local context and relevant comparisons.

Radar's approach represents a specific model of human-AI collaboration: Every story begins with human editorial judgment: what angle to take, what context to provide, how to frame the data for maximum local relevance. The journalist writes the template, establishes the tone, and makes the editorial decisions that shape how the story will read across all its iterations.

The NLG technology then handles data entry and regional adaptation. It plugs in the specific numbers for each area, adjusts geographic references, and ensures the story makes sense for its intended audience. But the system executes editorial decisions rather than making them.

Radar evolved from a private company called Urbs Media exploring NLG for London borough news in 2015. It became a formal partnership with PA Media in 2017, funded by Google's Digital News Initiative.

Editor-in-Chief Jack Lefley outlined PA Media's broader context: "We serve every UK and Ireland media company" with content spanning text, images, video, and live feeds. The agency operates on being "fast, fair, accurate."

The initial rollout was free to local publishers, serving as crucial market research to demonstrate demand before building a subscription business. By 2018, they had launched their news wire service, allowing local publishers to subscribe to specific geographic areas.

The fact that publishers are willing to pay proves that the service works, especially since local publishers are generally speaking notoriously cost-conscious and skeptical of new services. Some primary customers are major publishing conglomerates who pay based on geographic coverage area and audience size.

The team operates fully distributed across the UK. This structure gives them better insight into local variations and concerns while keeping costs low, which is essential for their subscription pricing model.

The geographic distribution highlights one of Radar's biggest technical challenges: the UK's complex administrative structure. Kelly outlined the complexity: multiple council types with upper and lower tiers, different boundaries for police, NHS, and fire services, and variations between Scotland, Wales, and England in how data is structured and responsibilities are distributed.

While Radar has built its success on traditional NLG programming, PA Media is developing generative AI projects that extend the automation model into new territory.

Martin Ashplant, PA Media's Product Development and Operations Director, outlined their approach to audio content creation: "We're building LLM-supported summarization for radio content" that can integrate Radar's local stories for regional radio stations. The system includes voice synthesis partnerships with companies like ElevenLabs, with regional customization options.

Crucially, the development maintains human oversight. "Customer control requires human approval for all AI-generated content," Ashplant emphasized.

The team acknowledges current limitations. Kelly noted that their response capabilities are "limited to national sources due to scope" and they "can't match the speed of in-house data journalists" for breaking local news.

Radar has attempted expansion into Germany and Australia with limited success, facing cultural resistance and fragmented media ownership structures that contrast sharply with the UK's consolidated market. The UK's unique combination of reliable government data infrastructure, consolidated media ownership, and established news agency relationships creates conditions that may be difficult to replicate elsewhere.

Radar has built a system that multiplies the impact of human journalism. One well-reported story becomes hundreds of locally relevant pieces.

The subscription model suggests genuine demand from local publishers for this kind of scaled, data-driven content. In an industry struggling with resource constraints, Radar offers a way to maintain local relevance without the overhead of dedicated data journalism teams.

Radar demonstrates a sustainable model for human-AI collaboration in local journalism, building a business around making journalists more productive and their work more valuable to local communities.

The automation serves journalism's core mission: getting relevant information to communities that need it. The technology handles the repetitive work of adaptation and localization, while human journalists handle the work that requires judgment, analysis, and editorial skill.

Radar's success offers several practical insights for other news organizations exploring automation: