What our subscribers (and a bit of AI) taught us about churn
Malaysiakini, one of Malaysia's leading independent news sites founded in 1999, is building Re-engage, a tool that uses predictive analytics to reduce subscriber churn
By: Andrew Ong, Malaysiakini
When we signed up for the JournalismAI Innovation Challenge, supported by the Google News Initiative we were under the impression that our churn problem largely had to do with content. Six months later, we've learned that it's just as much about engagement and payment behaviour.
Malaysiakini pioneered reader revenue in Southeast Asia in 2003. It was a long and painful journey before reader revenue finally became a substantial chunk of our income during the COVID-19 years.
In response to dwindling network advertising income, we tightened the paywall, improved checkout flows, reduced friction and experimented with offerings. The COVID-19 years also gave us four Prime Ministers within three years. This tumultuous period gave us our own version of the "Trump bump" that helped drive subscriber growth.
Since then, as politics settled and news consumption changed, we have experienced unprecedented levels of high churn. The subscription team wasn't short of ideas. They tried every way they could to market our product to former subscribers and those approaching expiry. There were broad campaigns (Press Freedom Day, National Day and more). Very reluctantly, they experimented with discounts. But what they were short of was data, specifically who are those considered a churn-risk.
Thanks to the support of JournalismAI, we set out to answer that question.
Different models for different readers
So far, we’ve built separate models for one-off and recurring subscribers. With the help of a data scientist (we’ve never worked with one before), what we found was vastly different behaviours, stickiness, and churn probabilities.
Why do we even offer one-off payments? In Malaysia, people generally like to make online payments through their bank accounts. This is a challenge because banks do not allow for recurring payments such as through Stripe. Asking them to set up recurring payments online would involve our subscription team being familiar with dozens of banking apps.
Hence it made sense to have separate models for both types of customers.
Our key findings:
People who manually renew make a conscious decision every year. However, recurring subscribers often leave for a variety of reasons, including failed payments.
In our journey so far, we discovered the following:
1. One-off subscribers who haven't visited for 30 days have about a 95% chance of lapsing.
2. Recent engagement matters far more than long-term trends. This indicates that if we have good stories, we need to reach out to churn-risk to re-engage them and help them build up a habit.
3. More importantly, our biggest retention opportunity may be converting people from manual renewal to recurring payment systems. We have some ideas and we will experiment.
In the coming weeks, we will be experimenting with our EDMs to those we now have on the watchlist. Among others, we will be testing some of the theories that we have come up with such as whether or not targeting those who have less activity with us or whether packages they have chosen will maximise impact. We will also have to set up a system that creates a new watchlist because the data would change after successful interventions.
In parting, we think that a good data scientist is worth the investment. It has been an eye-opening experience. Translating machine learning into newsroom language takes time. But with some patience (and perhaps a beer, or two), the conversations become a bit easier.
Another lesson so far is that churn isn't necessarily an editorial problem waiting for an editorial solution. Before experimenting with AI-generated recommendations or personalisation, it may be worth understanding your subscription data first. That foundation has already changed how we think about retention, and we're only halfway through the project.
If you are thinking about churn, do reach out. See you in the next post!
_____
This article is part of a series providing updates from 12 grantees on JournalismAI’s Innovation Challenge, supported by the Google News Initiative’s second cohort. Click here to read other articles from all our grantees.