Person with big shoes and with one hand on hip and the other pointing to a graph with an arrow going up.
Illustration by Elina Cecilia Giglio

How machine learning can help sustain local news

At Brown’s Local News Lab we are using Machine Learning to “optimize the ask”

Hannah Wise
6 min readNov 24, 2020


Hello from the Local News Lab at The Brown Institute! Since our last post, we have been busy connecting with a wide variety of newsrooms from the west side of Manhattan to Harare, Zimbabwe. Our machine learning engineers have been working on modelling and as a team we are planning some discrete experiments that will help us establish the direction of our first product.

Machine Learning (ML) and Artificial Intelligence (AI) are both very buzzy terms that have been floating around tech spaces for decades and journalism more recently, but this post is for anyone who is fairly new to the concept of ML and would like to know more about how it might help local newsrooms and their financial sustainability. Mark Hansen, the Director of the Brown Institute says:

“Journalism is becoming a computational exercise. Of course, computation is an important tool in reporting, and we make a host of data-informed decisions about our content right down to its placement on the website. But through computation we have also experimented with story automation and have created a host of fresh ‘news products.’ We have even used advanced ML to assess the diversity of the subjects we report on and the sources we turn to. But these all rally computation to create experiences for our readers. Now at Brown, we are exploring how computation opens up creative new opportunities to invite readers to become subscribers, members or donors.”

Person with long hair posing a question and speaking with ellipses in a speech bubble.

Umm…what is Machine Learning again?

If you are still a bit fuzzy on what ML is, here are two short videos that do a very good job at explaining it: this one and this other one.

ML is how Google and Facebook choose which ads to surface on the websites you browse. It’s how your favorite video streaming app makes recommendations on what to suggest you watch next. Michael Jordan, a leading ML researcher, defines the field as blending “ideas from statistics, computer science and many other disciplines… to design algorithms that process data, make predictions and help make decisions.” He sees ML as the basis for a new discipline of “Intelligent Infrastructure, whereby a web of computation, data and physical entities exists that makes human environments more supportive, interesting and safe.”

What does that mean for journalism? Certainly the people and processes behind the news make decisions that are more informed by data and computation, but it’s time to think about ML’s impact on the industry holistically — from content creation, to distribution and engagement, to business sustainability.

Person pointing to graphs and a technology stack.

Machine Learning is already all around us

Reader activity across news sites has been tracked and analyzed using some form of ML for years. The industry has adopted engagement metrics that summarize reader behaviors and provide benchmarks against which publishing strategies are ultimately evaluated. Recently, the platforms that publishers use to deliver news and other content have greatly diversified, and so too has the variety of ways we can now quantify a reader’s overall experience with a publisher.

With falling revenues from advertising, the reader (whether they be a donor, supporter, member or subscriber) has become a more critical and direct contributor to news sites’ revenue streams. ML is being applied to optimize business decisions in countless other industries — examples of Jordan’s Intelligent Infrastructures. At Brown, we are exploring how to make dynamic, personalized, and impactful choices about the forms in which we offer content to news readers, and the ways in which we ask for their support — or in much trendier (and pithy) words, to “optimize the ask”.

With the ability to adapt a publisher’s behaviors in real time, and uniquely for each reader, there are a myriad of possibilities for designing new experiences around content. Newsrooms, for example, have experimented with various kinds of recommendation strategies and personalized

feeds of stories and newsletters. Each of these are attempts to infer a reader’s interests by looking at their history with a site, or to predict content they would find useful given the activities of others who have engaged in similar behaviors. The latest ML tools are being used to optimize these individual recommendations, reader-by-reader. For instance, The Dallas Morning News has experimented with a separate paywall for their popular election voter guide that gave readers the option of choosing their own price to subscribe starting from $0.00.

Person with hair wrap typing on laptop while images of technology surround them.

Optimizing the ask

There is research that supports what many in the news business already can tell you they know from gut instinct and their own existing metrics: that readers who return to their site on a regular basis, and readers who engage at higher rates with content relevant to their location are more likely to subscribe. Many newsrooms around the world are already using data to influence their publishing, subscription and engagement strategies. By incorporating ML, we encourage cross-organizational discussions about readers, their interests and habits, and the data we have to represent our ideas. ML can help us think creatively about the ways we might structure our content to provide opportunities for an audience to support our business, while enabling experiments to see what works.

At Brown’s Local News Lab, we like to say our product will help get the right message in front of the right person at the right time — optimizing the ask. Because if we can learn what that sweet spot is — when and why a casual reader is most likely to hit the “subscribe” (or “join” or “donate”) button — newsrooms should see a boost in reader revenue, helping them on their path to financial sustainability, or whichever sort of engagement they wish to create. While “paywall” is indeed the easiest way to describe our product, it is actually more of a tool to facilitate conversion. Ultimately we envision a newsroom creating not just a reporting plan for a particular story, together with a publication and social media schedule for distribution, but also a conversion strategy — something that is structured and aligned with the content the newsroom is generating, and its publication.

While the initial allure of ML is that we can imagine new paywall functions and optimize their use, there is much more to the exercise. ML can help newsrooms better understand their business. What brings in subscriptions? Which content resonates? How should paywalls and other prompts like registration and newsletter subscriptions be deployed? Who is most likely to register or subscribe or even churn, and under which conditions and when? Unfortunately, not every newsroom has access to data scientists or machine learning engineers who can develop the technology to enable this level of insight.

This is where we come in.

In a nutshell, our goal is to help fill a pressing need in local news of high level data science that until now has been unattainable, and in doing so, deliver a product that uses sophisticated ML that will lead to increased revenue and a more financially sustainable product. We also have the experience to know the limits of ML — what’s plausible and what’s hype. Sorting fact from fiction is another service we can provide our local newsroom partners.

Please let us know if you have any questions about ML or our project. Feel free to drop me, Hannah Wise, News Partnerships Lead, a line.

And thank you for reading!



Hannah Wise

Product + Community Lead at the Local News Lab (@LocalAtBrown) at Brown Institute (@BrownInstitute) | Coach + Consultant | Former @cbcnews