Marketing Tech Grabs Headlines at Boston Globe
The 150-year-old media company uses MIT research to analyze and target audiences
By Peter Krass
As newspapers struggle to survive, some industry leaders are collaborating with academic researchers to develop new strategies and practices for the digital age.
Just ask Boston Globe Media. Over the last six years, this 150-year-old company has worked closely with the MIT Initiative on the Digital Economy (IDE) to keep pace with marketing technologies and industry pressures. As executives explained during the first CMO Summit@MIT, hosted in late April by the IDE, that research has resulted in improvements worth millions of dollars.
“The Boston Globe is certainly a success story in an industry that has seen tremendous challenges over the last two decades,” said IDE director Sinan Aral.
Market research has always been important in publishing, but the tools have changed dramatically. Working with the IDE to make data-based decisions, the Globe has overhauled its marketing strategy and pricing and has dissected its print and online audience in many new ways. “So much of what has gone into our strategy has been based on the learnings we’ve come to together over the years,” said Thomas Brown, Boston Globe Media’s VP of consumer revenue.
Part of the change was the appointment of the company’s first-ever CMO, Peggy Byrd, in late 2020. As the company transitions from home delivery to digital distribution, she told attendees, “the challenge is what to focus on, what to prioritize, how are you going to harness all that historical beauty.” Ultimately, she added, “our goal is to create this amazing consumer journey and attract people wherever they are.”
Reducing ‘churn’
Like many companies with a subscription-based business model, the Globe wants to lower the percentage of subscribers who fail to renew. Keeping this churn rate low is vital, because renewals equal revenue. By comparison, attracting new subscribers is more costly and less profitable.
One experiment conducted by the IDE team to address churn and retention involved subscription discounts and incentives to bump up renewals.
As Dean Eckles, MIT Sloan assistant professor and the IDE researcher who led the project explained, the campaign was more nuanced than it seemed. If Globe subscribers liked the discount enough to renew, that could help reduce churn. But if the discounts didn’t trigger more renewals, the discounts would lower revenue and hurt the Globe’s profits.
To help predict these long-term but hard-to-measure outcomes, Eckles and his colleagues conducted two experiments, he told CMO Summit attendees. In the first, they sent a single email offering a discount to 1,000 eligible subscribers who were judged at high risk of churn. The second intervention offered a wider set of discounts to a group of 6,000 readers. The team then monitored the subscribers’ retention rates over 18 months. (Read more in the research paper here and in this interview.)
But in the fast-paced world of digital marketing, “we didn’t want to be waiting long to learn from our experiment,” Eckles said. Instead, they used machine learning to analyze the first offer’s results and behaviors after the first 90 days, then used those findings to create a “surrogate index” to predict the results of the next intervention and adjust the recipient list for greatest impact.
The result? The short-term surrogate prediction was just as accurate as a prediction made after 18 months. Also, the team discovered that sending a discount to all subscribers actually had a worse payoff than sending no intervention at all. The key was picking the right subscribers — and that’s where machine learning helped the most.
Eckles estimates that these experiments helped the Globe realize a revenue increase of $4 million to $5 million over a three-year period.
Now, the Globe is applying these churn-management techniques to its half-a-million print readers and 200,000 digital subscribers. “It’s nice that we’ve had some really big wins,” Brown said.
Predicting preferences
Globe Media also worked with IDE Fellow Paramveer Dhillon to better identify audience interests: do they like sports, business, politics? Dhillon led a project to help the company model how consumers’ preferences for news consumption change over time.
Dhillon’s team harvested clickstream data from the Globe site, then used machine learning to identify patterns and to generally optimize the user experience. The challenge, Dhillon explained, was in understanding the dynamics of users’ latent or hidden interests.
Machine learning and a neural network model were deployed again to track the preferences of roughly 500,000 unique users over nearly five years, for a total of some 5.6 million observations.
By analyzing headlines, the model discovered latent interests in 30 categories including local politics, vacations, entertainment, social issues, baseball and basketball.
To protect subscriber privacy, all data was encrypted and de-identified before being analyzed. Based on this content-preference data, dynamic interest categories were created — including “sports fanatics,” “political junkies” and “local/ex-pat.” Readers were assigned to these categories based on how much time they spent on various article types
Finally, the researchers aimed to predict which content people would read next. Compared with other AI models, the approach was more accurate. The team also could detect site-scanning bots that could have distorted the data.
Going forward, the IDE will examine Boston Globe’s expanding universe of newsletters. “We’re learning a lot,” Brown said, “and we have a bright future.”
CMO Byrd enthusiastically agrees that data analysis is helping her identify new audiences and platforms to explore. She noted that the Globe now publishes more than 30 newsletters, sponsors about 300 live and virtual events a year, and has just launched a radio show, “Black News Hour,” aimed at Black listeners.
“Doing deep dives,” not just on potential customers, but also current customers, is key to retention, Byrd says. “You think that the people who started subscribing 10 years ago are the same, but they’re not. People grow and change all the time.”
Contributing writer Peter Krass is a business and technology journalist.
