As news publishers continue looking at how to better connect with users, personalisation is a reoccurring theme. Whether they’ve already started or are thinking about it, personalisation is inevitable.

During Wednesday’s Webinar, “Personalisation: Connecting the Digital Dots,” Gannett, Mediahuis, and NZZ shared their personalisation journeys, learnings, and results.

Jodie Hopperton, INMA Product Initiative lead, explained that what often is referred to as personalisation is actually automation: “True personalisation is one-to-one personalisation, where it’s based on my needs. But, actually, automation we kind of cover in this as personalisation.”

There are two types of personalisation, she said: active and passive. With active personalisation, readers tell publishers what they want — what topics they are interested in, what sports teams they follow, etc. Passive personlisation uses algorithms and is based on behaviours and demographics, among other things.

The main reason to personalise, Hopperton said, is because readers are completely different. Companies that don’t personalise may miss out on opportunities to capture users. And the consumer market is accustomed to personalisation; services like Hulu, Netflix, and Spotify are delivering deeply personal experiences. “So, to some degree, maybe the next generations or some generations now even expect it. So, the cost of not personalising could be high,” she said.

Gannett shares its personalisation discoveries  

Nicole Dingess, vice president, product design and UX/consumer product at Gannett, explained that personalisation is one of three core pillars of the company’s overall strategic product focus.

“We think of our product experience evolving, being more personalised to the user or to different cohorts, as well as being part of an immersive experience. We are looking at ways to get non-text content — the podcast, the video, the imagery — to really make you feel like you’re part of the event or where the action is.”

The other two pillars are being holistic solution that balances national with local coverage and being immersive, which means providing “content you can see, hear, and feel.”

Her team does a lot of concept testing around the idea of personalisation. This has included 10 major ethnographic research studies where they looked at target consumers in-depth, and they have had 1,900 users participate in user testing.

They also have a feedback loop on products in which 7,500 users have provided feedback on the USA Today and local sites. That has helped shape the company’s approach to personalisation. The target consumer is a younger, mobile-first user and they have learned a lot about what those consumers like — and what they don’t.

“We hear again and again, personalisation is expected and they feel like if they are using a prominent brand, the content should be based on their previous choices or what they’ve actively told us,” Digness said. “They’re looking for that frictionless experience and a product that they feel that knows them. So if they said they’re not interested in a topic, if they’ve already subscribed to a newsletter, don’t be resurfacing these things again because that makes them feel like you don’t know them.”

Active personalisation — the ability to actively curate the reading experience by filtering or saving certain types of content — continues to be important to users.

Another big need from younger consumers is the idea of cross-platform personalisation, Dingess said. “If I save something in my app experience, I very much would like that to be on my Web experience.”

Overall, consumers want products to be smarter and to know what they want to see next. That has led Gannett into rolling out such concepts as a subscription product that has a “For You” view that allows them to access premium content as well as additional modules on a single tab. “As a result of aggregating this into a For You, we saw about a 60% lift in engaging with subscriber content per user over the anonymous users,” she said.

Gannett has also released features such as the ability for users to text questions to a reporter and extending the commenting experience to make it more dynamic. And, for those who prefer to have a story read aloud to them, Gannett has added that feature, too.

Mediahuis is testing personalisation

Riske Betten, product director at Mediahuis in the Netherlands, shared a personalisation test it conducted for its Telegraaf.nl brand. The simple test looked at the homepage and selected three positions, then conducted an A/B test with two groups. One group saw articles selected by the editors and the other group saw the articles recommended by the algorithm. “The success criteria for now was the CTR of the position,” she said. Later in the test, they added video to the algorithm as well.

After conducting the test, they discovered personalised articles generated a significantly higher CTR compared to handpicked articles, with an increase of 23%. Betten said while this is similar to results seen at other publishing companies, “I think every product thinker is blown away by a CTR increase of 23%.”

The way the algorithm selected articles was similar to the way humans selected them, Betten said, but there was a notable difference in where those articles were positioned. “It’s really stunning to see that the algorithm is more personal. It tends to select articles that really ... set people to click.”

Editors, she said, tended to choose more news stories than the algorithm, so using the algorithm helped them have a better range of articles.

“And for us, it was very interesting to see that video gained some show, which is very important, I think. For all of us, video and also audio, we are still trying to figure out how it fits in our platforms best, but maybe the algorithm can help us here.”

She said one important learning was that 90% of users didn’t want video. “The algorithm, of course, can perfectly find this 10% of people that do want to consume video and can serve them even more.”

Based on these results, Betten said it appears the algorithm provides value as Mediahuis develops its personalisation strategy. “From the brand perspective, it gives us the opportunity to serve way more of our content,” she said.

NZZ focuses on algorithm-driven sections

Cristina Kadar and Sabrina Peterer of NZZ’s data and UX division, took Webinar attendees inside their journey to personalisation using an article page.

The motivation to create a more algorithmic-driven section was fueled by the COVID-19 pandemic and the interest in journalism that came with it, Peterer said. In a year of unprecedented growth, NZZ began looking at ways to engage and retain users in 2021 and beyond. To do that, they looked at how to keep them as subscribers.

“We looked at different areas within the product and also at core competencies that we already had within our company but had not leveraged so far,” Peterer said.  From there, they decided to focus on the newspaper’s next-read sections. That was a logical place to start, she said, because “we wanted something that is more contextualised, meaning an article, more of a journey, that is really different [depending on] where the user is coming from.”

When they began, sections had editorial picks on top that editors manually selected. An externally generated feed had ads and other stories, and has a very low CTR and completion rate. They decided that was the area to focus on.

“The goal was really to substantially increased engagement by offering articles that are really relevant for these users,” Peterer said. A second goal was to decrease the manual work for editors and free up their time.

“And we also wanted to leverage the first-party data we had collected and also the data science expertise that we had in-house. We wanted to leverage and also expand further by really having data feeds.”

Kadar, who is product owner of the data science and machine learning department, explained how they set out to achieve those goals. The news feed was redesigned, and the personalised feed was based on the long-term user history.

“So for logged-in users and otherwise for anonymous users, we would have a default feed using most-read signals. Four months after the first deployment there was a whopping 40% uplift in the click-through rate in the weekly click-through rate.”

Even more interestingly, that effect continued after the novelty and newness of the feed wore off.

For the entire section, this not only improved the click-through rate, but NZZ also saw an improvement in engagement. This was measured by the completion ratio, or the percentage of articles being read until the end after the user has opened them, Kadar said: “And there, if we compare our internal personalised feed versus the externally provided one, we had an uplift of 63%.”

She said they use one-to-one personalisation and compute the recommendations for every user every couple of hours: “We do that in a passive way, leveraging the profile of the logged-in users, so the user sees articles similar to what they have read in the past.”

Moving forward, NZZ will continue innovating and iterating and will create dedicated experiences for users, Peterer said: “We have a lot of ideas on certain experiences that we can improve depending on user groups, depending on devices the users are coming with or the usage context.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .