By Fiona Chan
Head, Media Strategy and Analytics, Singapore Press Holdings
July 23, 2019
Newspapers used to be the very definition of mass media - a one-size-fits-all, once-a-day product delivered to hundreds of thousands of households regardless of what each of them actually wanted to read and when they wanted to read it.
Digital media and machine learning has now turned this situation on its head. The Straits Times (ST) first launched its digital edition more than 10 years ago, serving up breaking news updates throughout the day. Users of the ST app can already curate their own push notifications depending on the content categories they like to read. Most recently, ST has been experimenting with personalised content recommendations, using artificial intelligence (AI) to display a customised set of stories and videos that each user is most likely to engage with.
This customer-first, data-driven approach takes in dozens of signals about every single user, from the device and operating system they are using to their frequency of visits and the articles they have read before. Acting on this information in real time, our AI models instantly sift through thousands of articles and identify the most optimal ones to display to each reader based on their unique attributes and reading behaviour.
One manifestation of this is that on the ST homepage, users see a selection of stories that are customised picks for their own tastes. But the influence of editorial judgement is still paramount - the top stories on the homepage remain hand-picked by editors to display for all readers, based on what the newsroom believes is the most important or impactful editorial content. This allows readers to continue to keep up with the biggest news of the day, while also enjoying targeted articles and videos that cater to their individual interests.
Users who are reading any article on ST are also served a list of other articles they might be interested in. These article-level recommendations are selected based on their relevance to the current article being read, as well as each user’s preferences and reading history.
Personalising readers’ experiences on our digital properties helps users discover content that is most relevant to their interests, and exposes them to the full breadth of content that our products offer. ST, for instance, publishes about 200 new articles a day. But readers view more than 30,000 different ST articles each day on average, indicating that tens of thousands of older articles have a much longer shelf life and continue to be interesting for our readers.
While the tailored content on the homepage is currently powered by Cxense, a machine-learning solution for personalised customer journeys that SPH is partnering with, some of the article-level content recommendations are the result of a custom-built AI engine. Together with local machine-learning startup DC Frontiers, SPH has been working to develop its own custom recommendation engine that we can have complete control over. This will allow us, for example, to adjust levers so we can showcase more editorially significant stories.
Since the personalised homepages were implemented, reader engagement has shot up. Click-through rates on the homepages have doubled, leading to a significant improvement in engagement metrics such as bounce rate.
On the article level, tests have shown that our custom-built content recommendation engine is particularly effective at engaging new users who are visiting ST for the first time. The engine is also surfacing a greater variety of articles with different freshness levels, compared to other engines which mostly recommend articles published within the past 2 days.
Reader engagement is one of SPH’s top priorities as we continuously strive to improve the user experience on our digital properties. An engaged user is a loyal user, which translates into more time spent on our content sites and more pages read per visitor - in turn making our users more likely to subscribe to our products and to be receptive to the ads we run.
As SPH continues to build on its audience-first mission, hyper-personalisation and catering to individual user preferences will remain key prongs of our strategy to build engagement and loyalty among our readers.
Fiona Chan is Head of Media Strategy and Analytics at Singapore Press Holdings, where she oversees business strategy, strategic partnerships, innovation management, data analytics and research for the media business. She was previously Managing Editor of The Straits Times and has spent a decade in business and political journalism.
A graduate from the Wharton School of the University of Pennsylvania and Harvard University, she was formerly an investment banker at Bank of America Merrill Lynch. Fiona is also a board member of the Land Transport Authority, Singapore Media Exchange, AsiaOne, and non-profits Sing Lit Station and The Straits Times School Pocket Money Fund.