It has become increasingly fashionable, and understandably wise, for digital subscription businesses to employ propensity models as cornerstones of their marketing approaches. When executed correctly, these models better identify segments of the overall audience more likely to convert into a paying subscriber, or conversely, cancel a paid subscription.
The attraction of these models to marketers is almost never more apparent or valuable than in paid media efforts. They can prevent the costly, unnecessary deployment of paid media to site users and visitors who have little to no interest in subscribing. Additionally, why spend precious marketing dollars on re-engaging subscribers who are already using their subscription actively sans encouragement?
At tronc, we have tens of millions of monthly unique visitors. We don’t have the luxury of being able to deploy paid media and remarketing efforts to all of them. To help guide our targeting and allocation of paid media, we conducted an analysis which essentially answered the following questions:
- Which Web site behaviours suggest a higher likelihood for a user to convert?
- What content typically attracts users more likely to subscribe?
- Are there any attributes, demographic or otherwise, common to users who subscribe?
We found the conversion of site users into paid subscribers spiked for site visitors who used the Web site search function or consumed local news content. Additionally, we identified the inflection points in engagement metrics, whereby once a certain threshold was exceeded, the tendency to convert greatly increased.
For instance, during Q1 for the Chicago Tribune, the conversion rate of users who visited the site five or more times in the prior 30 days more than doubled when compared to users who visited the site three times or less in the prior 30 days. Therefore, by targeting our paid media to those users, we concentrate our paid media to an audience twice as likely to convert, greatly improving our return on ad spend (ROAS) and reducing the “shrinkage” on our marketing dollars.
Common sense would suggest our titles are a more compelling proposition to Web site visitors located inside the designated marketing area (DMA). It turns out they’re almost seven times more compelling, based on the near seven-fold increase in the conversion rate of “in-DMA” audiences versus “out-DMA” audiences (Chicago Tribune, Q1 2018).
So with the analysis complete, we moved onto executing on the insights.
Propensity models that can’t be plugged into mainstream digital marketing systems such as Google AdWords, Facebook Business Manager, messaging or retargeting platforms, DMPs, CDPs, ESPs, or any other acronym-based ad tech may prove little more than academic exercise.
Thankfully, many of today’s digital ad solutions provide an “audience creation” functionality, allowing marketers to isolate users displaying behaviours that suggest a greater tendency to purchase, and then identify those users on external Web sites and apps using a range of ad products.
At tronc, we’ve built many of these higher converting audiences in Google Analytics and connected them to our AdWords account so we can target search and display ads to them. We’ve done something similar with Facebook Business Manager, the social network’s ad management console.
This simple marketing stack has realised a reduction on paid media cost per acquisition by 27%-45% for markets in our tronc portfolio.
We’ve also completed a churn risk regression analysis identifying the core behaviours and attributes that indicate a paying subscriber has a higher likelihood to churn. For instance, 81% of voluntary stops are from subscribers who’ve made fewer than three payments. And the risk of cancellation for subscribers who haven’t visited the site within the last 22 days is 133% higher than those who have.
Understanding those thresholds means we don’t need to port our entire subscriber base into a Facebook custom audience and target paid content ads to the entire base. With a simple SQL script and a marketing automation vendor, we can focus our paid media engagement efforts on Facebook to only the “high-churn risk” subscribers, managing our content re-engagement expenses and reducing voluntary churn more cost effectively.
None of these, by any means, is an overly complex or expensive marketing approach. The next chapters in our data-driven marketing journey will see us create more sophisticated, robust predictive models for acquisition and retention. These models will consider many different user attributes simultaneously to construct and assign site users with propensity to subscribe or churn scores.
However, the marketing efficiencies we’ve generated from these early-stage, easy-to-execute predicative models is encouraging validation for our data-driven approach and is energising the organisation’s push to continue and deepen this form of marketing.