Over the past six months, the South China Morning Post (SCMP) collaborated with the London School of Economics, Google News Initiative (GNI), and 18 other media companies on an initiative called Journalism and Artificial Intelligence (JAI) to develop a loyalty playbook supported by Artificial Intelligence (AI) strategies.
In a report that I co-authored with JAI, How can AI help build audience engagement and loyalty?, we learned publishers measure loyalty in the context of retention and churn in very different ways.
Here are highlights of some best practices around data and actionable intelligence to improve reader retention and mitigate churn, as well as how AI can be an essential tool for any publisher.
Understanding two key variables: churn and retention
First, we must define churn and retention in the context of the media as a publisher.
Churn (i.e., the rate of attrition or customer churn) is the rate at which our customers stop doing business with a publication. Churn can also give us valuable insights into our customers. By deepening that understanding we can develop the most effective targeted value propositions, and by sharing the knowledge of churn across all functions, we enable coordination, leading to greater efficiency.
It can be defined differently for between organisations, so here are two examples of how a subscription and a broadcast business might define churn.
- Subscription model: A user who cancels their subscription and/or does not renew within a pre-specified window (e.g. 30 days) after their subscription expires or lapses.
- Broadcast/free/freemium model: A user who previously engaged with a product who no longer engages with any product within a pre-specified window (e.g. 14 days).
Retention is a multi-definition term that can differ depending on company and business application. Defining retention is critical as it shows your value proposition is connecting with the audience. It may also suggest augmented products to expand into.
Here are some questions that may be applicable to determine what retention means:
- What’s the desired behaviour for your customer (e.g. how many articles read per month, time spent on site, or actions taken)?
- What is the customer behaviour you want to avoid?
- What are the top 5% of your users doing (most engaged or highest revenue-generating) that you would like to replicate across a broader range of your users?
- What’s the average frequency with which your readers return? Segment or cluster users by behavioural patterns to identify any trends.
Action around data
Understanding the metrics around churn and retention is only the first step. The next, and arguably the most critical, is taking action on that data once it’s known. Once these have been tested and validated as effective, they should be automated as quickly as possible.
For example, churn risk scoring has become a crucial issue for business and for publishers dealing with retention and loyalty. Setting benchmarks around churn and performance can help to create AI flags to signal when performance is going awry.
Churn definitions differ across businesses, so rather than build one model or prediction algorithm, here are examples of leading indicators that can be considered relevant across many organisations and publishers. By defining churn and retention, these definitions can be clarified and distilled into a framework of loyalty, which typically includes several if not all of the following parameters:
- X number of articles read in last Y days.
- X number of visits in last Y days.
- X number of days since last visit.
- At least X time spent on site.
- Visitor exit points.
Case study: Employing an AI approach at SCMP
As news publishers, we want to maximise the impact of our content, attract new users, alert consumers to new content or offers, or improve distribution strategies that serve user needs at optimal times. There are useful ways of leveraging AI to achieve this.
Here is a case study that illustrates using metrics to better measure churn, retention, and loyalty though recency, frequency, and volume (RFV). Together, they define engagement and represent a technique that database analysts use to segment customer data. This works by plotting the activities and behaviours of a user against three key actions:
- Recency: When’s the last date/time that the user came to your site?
- Frequency: How many visits or sessions within the past 30 days?
- Volume: How many pageviews and how much time spent on site?
Scoring RFV on a per-user basis enables you to roll up user data by article, topic, section, geography, platform, or any other grouping you’d like and leverage this metric quite broadly.
As an abstracted index or aggregated metric containing several variables is complicated to decipher for the less data-savvy, this can be turned into a percentage and reframed as a quality user score. Our quality user percentage simply indicates the percentage of users who obtain a certain RFV threshold or higher.
The algorithm we created recalculates the score for each of our users automatically every day and stores a historical record of each user’s score. This way, we can track increases and decreases across users, topics, desks, and much more. It is important to compare across a baseline on how a type of content performs, and it is useful to understand changes in quality scores across the same topic or desk over time to better understand trends and how to regain users or engage them more meaningfully.
Ultimately, there is no silver bullet or magical algorithm that will solve churn or increase retention without hard work, testing, and iteration. Each publishers’ customer base is unique and every organisation has different goals.
Understanding and anticipating audience needs is one of the clear areas where AI can benefit a news media business and, in particular, the newsroom attempting to win new audiences and retain existing users or convert them to more engaged users.