How RADAR became front page news
Lessons from the first year of an automated news agency
We have been marking an anniversary this week at RADAR, the automated local news agency. It’s 12 months since we kicked off the project to turn an idea about using AI tools to enable the mass production of local stories into reality.
Over the summer we expanded the pilot phase of our fledgling service to an open free trial. In an intensive production push the team of five reporters has worked on 130 story projects, producing multiple local version for each. In just over three months we have published more than 40,000 individual local stories. These have been distributed to nearly all the local news publishers across the UK for use in print and digital products.
The first year has been a steep learning curve in inventing a new way of working for journalists and talking with publishers to develop an innovative news service. So, what have we learned? Lots, but first a bit of background for those not familiar with RADAR.
What is RADAR?
The business is a joint venture between data-journalism start-up Urbs Media and the Press Association, the UK national news agency.
Since 2014, Urbs had been building an experimental data-driven London news site, as well as writing data stories for national newspapers and broadcasters. We wanted to scale up our work for every local market.
RADAR grew out of the idea that the increasing amount of open data that was being made available by Government departments and other public bodies, such as the NHS, contained news stories that weren’t reaching every corner of the country. As much of the data was segmented down to local level each spreadsheet might contain hundreds of individual local stories.
To write all those stories would require an army of journalists, all working on the same set of data, unless you could do it centrally and harness technology to scale up the production.
We believed that we could achieve this by combining human journalism skills with Natural Language Generation, a technology that basically allows a sheet of numbers to be translated into a page of words.
The Press Association, with its central role in the UK’s news production and distribution system, was a natural partner. Together, we successfully pitched to the Google Digital News Initiation for funding to support the development. RADAR — Reporters and Data and Robots, was born. We began work in early September 2017.
Innovations still need old-fashioned skills
Our first task was to hire a team of reporters, perhaps a little ironic for a business which has been described as robot journalism.
Some of the skills we needed were obvious, and common to all newsrooms. We needed journalists with a great nose for a local news story and strong writing skills. They needed to be comfortable working with numbers and using Excel. But they would also need to adapt to a new discipline of writing their copy within a Natural Language Generation programme.
This is not easy. Basic training takes two weeks. Experience so far has been that it is around a month before people feel comfortable with the technology, and the job is challenging. A writer must consider how the numbers in each row of the data might affect each sentence. It is not writing one story but hundreds of stories simultaneously. The mental gymnastics required can make this a pretty intense writing experience.
Our brilliant team has a diverse skills background. Most have a good grounding in local journalism. Some have an educational background in mathematics.
Data journalism often places a great deal of emphasis on skills and tools. But the greatest skill in this job is the ability to see a good story.
Robot journalism is only as good as the humans operating it
In many industries there is mounting concern that AI will replace people. Machines are already capable of doing many workplace functions, but they are terrible at some of them.
An aim at the start of our project was to find a way for technology to amplify the work of a journalist, adding to what they could do, not taking over their tasks.
In looking at how a story is developed it is clear that humans are much better at many tasks than machines. Identifying the best datasets to produce a story that is newsworthy, revelatory or topical is best done by a journalist. So is finding that story in the dataset. And humans write better copy than computers.
But the ability to replicate the process across up to 400 areas could only be achieved over a long time, by many people, or by using technology.
Our journalists select the most promising data, mine the data to find the story, develop the different angles and then compose a template that instructs the technology on what sentence to write as it computes the numbers in the spread sheet. We are writing stories as mini-algorithms for each new set of data.
The difference in what we are doing compared to the automation of financial results or football scores is that 400 versions of a local news story cannot work as a one-size fits all model which broadly has three different schemes — up/down/unchanged or win/lose/draw. A local news data story may cover many different metrics and each version of the story must offer the most newsworthy data points in the intro. The best angle changes from area to area. It takes journalistic skill to determine this and considerable writing ability with NLG software to deliver it.
There’s no shortage of data and much of it can make front page news
An early concern that was raised with us when telling people about our ambition for RADAR was that there would be insufficient data to sustain the service. This has not been the case.
Thousands of sets of statistics are published by the Government alone. At the start, the RADAR agenda was driven just by new data releases and we can’t keep up with all of them. We have been building in more topicality, looking at the broader national news agenda and digging out the latest figures that can offer local insight.
We have been working with non-governmental bodies and private organisations who are happy to share data with us and make it open. We feel we are just scratching the surface as the amount of available data grows.
We started the project with the bold assumption that there was news gold in this mountain of unexplored data. But we were modest in assuming that many of the stories we found would be interesting but more middle of the book (in newspaper terminology) than front page fodder.
We have been delighted to find the reality is better than that. Many of our stories are page leads in print or front-page splashes.
Data journalism can be a daily service
RADAR is about scale. It’s about finding lots of stories for lots of publications and making them available quickly. This is why our focus has been open data not Freedom of Information or investigatory journalism.
Our aim is to deliver to our customers a daily diet of strong, verifiable, data-driven stories that inform their readers about their locality.
We cover the beats that are familiar to local news — health, crime, transport, education, housing, the environment and lifestyle. In each of these, data at local level provides an insight into our daily lives.
We are delivering a slice of this insight every day to add into the mix of local news being produced by the newsrooms.
RADAR is a ready to publish, agency service and many of our stories appear quickly online, as we sent them. Others are used as the basis for a story that is developed by a local team, adding layers of context and comment from their area.
A new service can present a challenge as well as a benefit for customers
From the outset we have worked with our customers to develop the service. We engaged all the major local news publishers plus some independents and hyper-locals and set up a pilot group of about 50 titles. Feedback from this pilot group about topics and usage helped us steer the first few months of the service.
A lot of work in the early months was dedicated to developing a distribution platform. As we are writing hundreds of versions of each story we couldn’t send them all out via a single wire feed.
We map all data stories onto the local authority structure of the UK. This enables a user to log on to our platform, select a single area (or multiple areas) and see all the stories related to the area. This may include data stories about the local authority or its corresponding health services, police force, fire and rescue, schools, courts, prisons etc.
We are working with clients on how they handle the service, as for the larger organisations it means thousands of additional stories each week. We want this to be as frictionless as possible for them, delivering a new stream of content that they would not have been able to do, with the minimum call on their busy staff. Some have used their existing centralised content operations to select and distribute the stories, while others have left it to the staff at their various titles to choose and handle individual local versions.
We have been looking closely at traffic patterns on the platform to determine the best times and days for us to post stories.
The co-operation and feedback of all those taking part in the pilot and current free trial period has been invaluable.
The impact of the service
Tracking the usage of the thousands of stories we produce has been a challenge for us. We are getting better at it and are pleased at what we see in terms of usage and engagement.
The current Cairncross review into the sustainability of high quality journalism will mean a focus on the role, health and prospects for local media. We believe that RADAR delivers in a number of areas.
In scrutinising a vast array of local data we aim to provide stories that help citizens to be better informed about their areas and the decisions being taken — a societal benefit.
At a time of accusations of fakery and mistrust in the media we believe that a strong stream of transparently sourced stories has real value — a content benefit.
We are writing stories that previously would not have been covered by most publications as they do not have the time, staff or skills to do this work. It is a new stream of content for hard-pressed businesses to monetise — an industry benefit.
Contrary to the assumption that AI will take our jobs we are using the technology to develop new areas of journalism and explore how the role of reporter may change — an innovation benefit.
Next Steps
So that was year one. What next?
We are continuing to develop the service. We are just putting the finishing touches to the process which will enable us to generate bespoke graphics with each individual story — this will start with some bar and line charts in next few weeks.
Then we will be looking at automated video to complement some of our stories, each one locally tailored.
The support of the Google DNI has enabled us to create the RADAR service and test it across the local news industry. We have been delivering stories into the UK market for 10 months without charge, through the pilot phase and the current free trial.
The free trial period ends in the coming weeks. Given the positive feedback we received and the extensive use of the content, we hope that all the organisations who have been working with us to develop the service will be continuing with us as customers in the months ahead.
Our longer-term ambition for RADAR extends beyond the UK. As global pioneers in bringing AI to the local news industry we have attracted a good deal of attention, so there have been approaches from potential partners in many markets.
While our primary goal will be to establish a firm position in the UK market, we hope that year two will see us taking our brand of local journalism to a global audience.
