The South China Morning Post (SCMP) is an English-language newspaper based in Hong Kong, where it has been the city’s newspaper of record for 118 years.

In 2016, we pivoted to serving a worldwide audience with comprehensive and insightful China coverage — and the majority of our readership is now based overseas. Today, SCMP is the platform of choice for informed discourse on China and its impact on the world.

To meet the needs of our diverse global readership, we’ve taken a data-first approach to designing our user experience.

Creating a seamless experience for anyone that comes to the Post is essential. Personalisation helps us gain a better understanding of our readers, who come from a variety of backgrounds and seek out our content for different reasons. Prediction models ensure we can do this consistently.

Curation and engagement are vital to our platform. We realised we needed to declutter the user experience and streamline the call to action so that we could put content — the key driver for conversion — front and centre. We extensively studied ways to leverage data so we could create better user journeys on our platform, and we have learned a lot through our digital transformation.

This transformation involved asking ourselves how we could mobilise machine learning and data to develop a comprehensive and fulfilling reader experience that effectively guides users toward conversion without compromising effectiveness by “asking too much.” Too many prompts to sign up for a newsletter, watch a video, or create an account would detract from our main value proposition, which is our world-class journalism.

We recognised there are steps readers take that make them more likely to subscribe and less likely to churn. Identifying these steps would be crucial to driving engagement while prioritising our journalism. But because of the diversity of our readership, we could not work with a uniform path to conversion. In this regard, segmentation is key to understanding user behaviour.

If we could detect the type of user and identify the best action they could take, we could build a product that consistently drives engagement, conversion, and retention. With this in mind, we implemented a framework that categorises users according to four factors:

1. Location.
2. Behaviour.
3. Loyalty.
4. Persona.

From here, we can identify segments based on combinations of these factors and cast a broad net based on the region or precision-target specific segments.

For example, we can focus on regular visitors from the United States with an interest in lifestyle content who have not signed up for a newsletter, or we can simply focus on all readers located in Hong Kong.

We determine personas for our readers by analysing their content consumption patterns. We have several user personas, such as:

• Hong Kong readers who want local news.
• China watchers with an interest in in-depth analysis on policy and economics.
• Lifestyle enthusiasts exploring fashion and wellness content.

Each group has different levels of engagement and rates of conversion. Therefore, we tailor the user journey accordingly.

For instance, with lifestyle content seekers, we focus on enhancing discovery by nurturing their interests and increasing serendipity with recommendations, while suggesting they explore new interests. Hong Kong readers, on the other hand, have a more immediate need for our news coverage, so we can encourage them to subscribe straight away.

Once we’ve identified the user segment, we define the next best action. We can draw on our marketing and product expertise to analyse the effect of anything a reader decides to do on their lifetime value, such as sharing a story, reading a special report, or paying for a subscription. It is equally important to predict the likelihood that a user would take any of these actions.

To this end, we developed a priority matrix, which allows marketing teams to personalise campaigns in real time. Depending on the visitor’s loyalty, we know which action is most likely to lead toward subscription, which ranges from flyby to casual reader to paid subscriber, and their user persona. We built this framework by creating propensity models that specifically focus on conversion and churn.

When we detect a loyal user is likely to subscribe, we experiment with different messaging, paywall metres, and marketing e-mails. Similarly, we tailor the subscriber experience to ensure the user maintains their level of engagement; some respond well when we suggest topics to follow, while others are engaged by deep-dive content, such as podcasts. For those who are least likely to subscribe, discounts are the best incentive.

We have more than 20 user segments, which we use to plan campaigns with a personalised user journey building up to the call to action. We see much better results when we can configure the on-boarding flow in a way that effectively engages every segment.

Crucially, the framework also helps us ensure our recommendations are relevant to individual users, so that every step of the journey matters, whether they subscribe or not. Time spent on our platform should always be meaningful.

We’ve learned that our framework is most effective on the middle and bottom of the funnel when we have more information about our target users. The more we know about a user, the better we can serve and engage them.

By assessing the accuracy of our models, we’ve also come to understand that we have to target the right size of audience segment with personalised user experiences. If we target a group too large, the approach becomes generic and dilutes the effect. Focusing on a group that is too narrow makes a negligible difference.

Implementation, integration, and scaling all present challenges. It can take a while for marketing teams to adopt these models in their workflow, because it requires a change of mindset. Ensuring that the resources are accessible across systems involves troubleshooting. Considering their complexity, prediction models require maintenance and upgrades to ensure errors are avoided and identified quickly if they do occur.

We take an experimental and iterative approach so we can improve functionality and expand the scope of its application. These tools work, and they empower product and marketing while enhancing user experience. Given their complexity, they can be challenging to implement across an organisation. But they drive results at a scale that makes it well worth the investment.

Developing prediction models is increasingly important in the news business as we seek to engage readers over the long term. It generates a steady stream of traffic and helps provide a more well-rounded user experience during news cycles when certain topics dominate the front page over an extended period.

Personalisation can be applied across the business, enhancing workflows in marketing, audience growth, and product development. Understanding individual users matters more than ever before, and prediction models provide an intelligent approach.

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