When I first joined the South China Morning Post (SCMP) team a little over three years ago, our data team was asked to compile personas with limited resources and no data infrastructure in place. Eager to help, we cobbled together a few user interviews and some demographic data from Google Analytics into a report. However, with only a sample of user data and limited behavioural information, the analysis was perhaps more interesting than actionable.

Fast forward to the present day, and I’m proud to say we’ve come a long way in building up our data infrastructure. We’ve also further enhanced our methodology, enabling us to better understand our readers. We employed machine learning to assist with the enhancement of user profiles; implemented surveys, quizzes, and polls to deepen our understanding of reader preferences and behaviour; and invested heavily in incentivising users to log in so we’re not wholly reliant on the fleeting cookie.

All these efforts over the past three years have helped enable us to create these personas, which will hopefully help us better serve our readers.

Personas are representations of who our readers are, an archetypal description and humanistic story around an individual customer. They don’t simply illustrate what readers do or how they consume products, but also why. Good personas provide context on who our readers are and enable teams to take actions to improve engagement.

Rather than group users by geography or pre-defined segments, we wanted to use unsupervised learning to cluster users based on behaviours, regardless of referring source, platform, or geography. To do this, we used k-means clustering to group our users into cohorts and ran about 40 iterations to observe the optimal turning point in the WCSS (Within Cluster Sum of Squares).

This clustering yielded six personas with varying degrees of engagement, subscription potential, and interests. The sizing of these clusters also varied substantially.

We then spent several weeks running test campaigns — A/B tests to validate if one cohort outperformed another in propensity to subscribe to newsletters or registered to validate engagement levels and subscription propensity.

Following this, we interviewed dozens of users across each persona to validate the quantitative data with qualitative data. These conversations enabled us to supplement our quantitative assumptions with anecdotes, reader insights about what they wanted to see, feedback on product improvements, content mix, price sensitivity, and much more.

Some of the questions we asked included the following:

• What are our users’ goals and motivations for visiting SCMP?
• How did they find us? Through which touchpoints, channels, or funnels did they come through?
• How often do they visit SCMP?
• What’s their content format preference (bite-size, long-form, video, podcast, etc.)?
• Who are they (age, gender, industry, etc.)?
• What types of content (topic, section, etc.) do they consume?
• Where do they come from geographically?
• When do they usually visit SCMP (time of day, day of the week, etc.)?

As we built these personas over several months, we knew the most important output of this project was to communicate and share these personas across the company. We wanted to empower various key stakeholders to make these personas actionable and integrate the usage of them into our daily workflows.

• Marketing: Launch acquisition, retargeting, and retention campaigns to increase user engagement with custom messaging for specific personas.
• Product: Customise user funnels and experiences per persona so our users can have a more relevant experience and customise content recommendations for individual personas.
• Advertising: Enhance personas further with first-party data, and provide partners and advertisers with deeper insights on who their potential customers could be. Enable more targeted ads, higher conversion rates, and a more relevant ad experience for our readers.
• Editorial: Develop content around specific persona types and collaborate with product to customise experiences.

There is still much work to be done on the implementation of personas above, and we are excited to see how personas can help us more meaningfully engage with our readers.

Banner image courtesy of Pixabay.

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