Propensity Modelling: How publishers ensure their subscription business keeps growing
Data. Everybody’s talking about it, but how many publishers are walking the walk?
At this year’s Digital Publishing Awards, The Independent won Best Use of Data. And what set their data strategy apart? The use of propensity modelling. Modelling which saw their readership figures grow to an impressive 98 million readers per month.
In an age where data is often cited as the most valuable asset around, the use of propensity modelling to learn more about your customers cannot be avoided. It’s a valuable tool for the modern digital publisher, and nothing short of essential if you want to truly understand your readers’ behaviors and leverage this information to maximize revenue.
And whilst market segmentation has long been used to target certain customer groups and promote certain behaviours, propensity modelling is the next step forward. But what exactly is a propensity model? How does it enable digital publishers to retain and build on their subscriber base? And how does it improve upon existing market segmentation strategies?
What is a Propensity Model?
A propensity model is a specific type of statistical model used to predict the likelihood of a certain behavior or outcome based on historical data. In the context of digital publishing, a propensity model assigns a score to individual users, indicating their probability of engaging in specific actions, such as subscribing, renewing, or unsubscribing.
Key Components of a Propensity Model:
- Data Inputs: The model utilizes various data points, including demographic information, past behaviors (like article views and engagement levels), and interaction history (such as email clicks or responses). The quality and breadth of this data significantly influence the model’s accuracy.
- Algorithms: Propensity models can be built using different algorithms, ranging from logistic regression to machine learning techniques such as decision trees or neural networks. The choice of algorithm often depends on the complexity of the data and the specific outcomes being predicted.
- Score Assignment: After processing the input data, the model generates a propensity score for each user. This score quantifies the likelihood of the user engaging in a targeted behavior—higher scores indicate a greater likelihood of action.
What is propensity modelling?
Propensity modelling uses data to predict the future behavior of customers. For example, this modelling can help you identify how likely a reader is to subscribe or churn, providing accurate foresight into future actions. Each reader is allocated an individual, personalised propensity score, making identification for the likelihood of a certain action simple- e.g. the likelihood of them converting.
How does it work?
Propensity modelling can take various forms. However, the main way it can be so accurately predictive is through Propensity Score Matching. The model looks at the behaviors of previous readers and customers and compares them to those of new ones. Their future behavior can be determined because groups with similar propensity scores mimic one another. If new readers share the same behavioral patterns and characteristic backgrounds as past ones, they are likely to act in the same way.
Why is propensity modelling useful for digital publishers?
Crucially, propensity modelling is a tool that can help digital publishers both convert new subscribers and retain existing ones.
On the conversion side, propensity modelling looks at the behavior of the reader and feeds them appropriate content and offers so that the likeliness of subscription is maximized. For example, a reader who is found to be unlikely to convert by the model could be sent a personalized promotional email with an offer to entice them. On the other hand, customers who are likely to convert and don’t need further persuasion should be shown an effective paywall straight away so that they don’t fall at the last mile. Hence, propensity modeling acts as a tool to ascertain the most effective methods to grow your customer base, increase revenue and scale your subscription business without decreasing your profit margin through unnecessary price cuts.
On the retention side, propensity modelling observes how engaged individuals are with your product. Modelling allows your company to discern which customers have a high propensity towards renewing their subscription and those that don’t. This allows you to maximize your time allocation and budget by focusing on genuine churn risks.
Propensity modelling enables your strategy to be targeted to the behaviour and needs of each individual reader – moving away from a one size fits all strategy. Whilst some readers will subscribe regardless of how many articles they read or however much targeted marketing they receive, others will require unique personalization.
In anticipating the actions of both known and unknown readers, propensity modelling takes the uncertainty out of strategic planning. With greater confidence in customer retention and conversion, investments can be made in high-ROI campaigns, maximizing revenues.
What types of propensity modelling can digital publishers take advantage of?
Propensity modelling can predict many differing aspects of your users’ behaviors. Here are some of the main types you need to be making use of in digital publishing:
Propensity to engage: The likelihood that a reader will click on any promotional material they receive.
Propensity to convert: The likelihood a reader will become a paying customer – guiding you in identifying which readers require more persuasion before converting, maximizing their chances of subscription.
Predicted lifetime value: The likelihood a reader remains subscribed for a long time and therefore, how valuable they are to your business over their lifetime. Marketing efforts should be concentrated on high-LTV customers as they will bring the most revenue in the long run, generating a larger return on investment. Your most loyal subscribers are likely to be your loudest advocates, too.
Propensity to churn: The likelihood that an active customer will unsubscribe from your business. Churn risks need to be identified as soon as possible and re-engaged if there’s an option to do so.
How does propensity modelling fit with segmentation?
The use of propensity modelling can help you take market segmentation to the next level. Traditionally, market segmentation involves splitting up your target market into groups with similar characteristics and behaviours, such as by age, geography, or user status. This is done under the assumption that customers who are similar in profile will act in a similar way and can thus be targeted by the same methods. The more data points you collect on an individual, the higher the level of similarity between customers can be gauged.
Propensity modelling elevates the segmentation strategy, providing an additional, critical, data point. Publishers need to consider not only who their readers are but also how they act and not just take for granted that similar people will act in similar ways. Segmentation is a valuable strategy and propensity modelling can propel it by adding a mathematical source of certainty.
For example, when targeting a group of 18–24 year olds from south London, use of blanket messaging that you believe applies to the group is an easy starting point. However, using propensity modelling to understand which individuals within the sub-group are genuinely likely to convert is far more valuable.
While propensity modelling does consider customer characteristics to generate propensity scores, segmentation can also support propensity modelling. For example, take two customers who both have a low conversion propensity score but one of them is from a high-income background and the other from a low-income one. The conversion strategies for these two readers will differ. While for the low-income reader conversion may require discounts, for the high-income reader it might be more effective to demonstrate premium value, such as making sure interest tailored stories pop-up for them first when they open the site
Challenges in implementing propensity models
While propensity modelling offers significant advantages for digital publishers, implementing it comes with its own set of challenges. Here are some common obstacles to consider:
- Data Quality and Availability: High-quality data is essential for effective propensity modelling. Publishers often struggle with incomplete, inconsistent, or outdated data, which can lead to inaccurate predictions. Ensuring that data is clean, comprehensive, and up-to-date is a crucial first step.
- Resource Constraints: Developing and maintaining propensity models requires dedicated resources, including skilled data analysts and robust technological infrastructure. Smaller publishers may find it challenging to allocate these resources effectively, potentially limiting their ability to leverage modelling.
- Complexity of Models: Propensity modelling can involve sophisticated algorithms and statistical methods. For teams without a strong data science background, the complexity can be overwhelming. This may lead to misinterpretations of the data or ineffective implementation of the models.
- Integration with Existing Systems: Successfully integrating propensity models into existing marketing and operational systems can be a logistical challenge. Publishers need to ensure that their CRM, marketing automation, and analytics tools can work seamlessly with the propensity models to realize their full potential.
- Privacy Concerns: With increasing regulations surrounding data privacy (e.g., GDPR, CCPA), publishers must navigate legal considerations when collecting and using customer data for propensity modelling. This requires a careful balance between personalization and compliance.
Best practices for successful propensity models
To maximize the effectiveness of propensity modelling, publishers should adopt the following best practices:
- Start with Clear Objectives: Define what you want to achieve with propensity modelling—whether it’s increasing conversions, reducing churn, or enhancing customer engagement. Clear goals will guide your model development and data collection efforts.
- Invest in Data Quality: Ensure your data is accurate, consistent, and relevant. Implement processes for regular data cleaning and validation, and consider investing in tools that enhance data quality.
- Segment Your Audience: Use propensity scores to segment your audience into meaningful groups. Tailor your marketing strategies based on these segments to increase relevancy and effectiveness.
- Continuously Monitor and Update Models: Propensity models should not be static. Regularly review and refine your models based on new data and changing user behaviors to maintain their accuracy and relevance.
- Leverage A/B Testing: Use A/B testing to validate your modelling assumptions and test different marketing strategies. This helps identify which approaches are most effective for different audience segments.
- Collaborate Across Teams: Foster collaboration between marketing, data analytics, and IT teams. A multidisciplinary approach can lead to more comprehensive insights and innovative strategies for using propensity modelling.
Key takeaways
Fundamentally, propensity modelling, coupled with segmentation, allows you to efficiently allocate resources in your acquisition and retention efforts.
It is essential for modern digital publishers to move beyond just segmentation if they are to maximize their revenues and retain a lively customer base. With so much data available which can be capitalized upon, predicting the future of your customers to an accurate degree is now a reality, and might prove essential to securing the future of your business.
A powerful subscription experience platform can take care of segmentation and future prediction. For example, user segments are built into Zephr’s subscription platform under the rules builder, allowing for simple, code-less personalization, while “Optimize” (Zephr’s new powerful analytics) monitors your customers’ behaviour to spot critical behaviours early.
FAQs About Propensity Models
Q: How do I start implementing propensity modelling?
A: Begin by defining your objectives, collecting high-quality data, and selecting the right tools or software. You may also consider collaborating with data analysts or data scientists for model development.
Q: What kind of data do I need for effective propensity modelling?
A: Key data types include demographic information, user behavior metrics (e.g., page views, time spent on site), and past purchase history. The more comprehensive your data, the more accurate your predictions will be.
Q: How often should I update my propensity models?
A: It’s recommended to review and update your models regularly—at least quarterly. This ensures they remain relevant as user behaviors and market conditions change.
Q: Can propensity modelling be used for more than just subscription services?
A: Yes, propensity modelling can be applied across various industries to predict customer behaviors, including e-commerce, retail, and service industries, enhancing marketing strategies and operational efficiencies.
Q: What are the potential ROI benefits of implementing propensity modelling?
A: By accurately predicting customer behaviors, publishers can enhance marketing effectiveness, reduce churn, improve customer retention, and ultimately drive revenue growth, leading to a higher return on investment.