Propensity Modelling for Digital Publishers

Marketers talk a lot about how important it is to deliver the right messages to the right people at the right time. Of course, it's not simple to pull off. That's especially true when you consider how many marketers still use a one-size-fits-all, spray-and-pray strategy to interact with prospects and consumers.Propensity modelling is one method marketers may employ to solve this barrier and achieve higher customisation and better business outcomes.

In an age where data is sometimes referred to as the most precious asset, using propensity modelling to understand more about your clients is unavoidable. It's an invaluable tool for the modern digital publisher, and it's a must-have if you want to properly understand your readers' habits and use that data to increase income.

Propensity modelling is the next step ahead from market segmentation, which has long been used to target certain client segments and promote specific behaviours. But what is it, exactly? What does it take for digital publishers to keep and grow their subscriber base? And how does it differ from conventional market segmentation methods?

What is a propensity model? 

8 Ways to Extract Value from Customer Data | Blog | Online Digital  Marketing Courses

Data is used in propensity modelling to forecast  customer behaviour in the future. This modelling, for example, may help you predict if a reader will subscribe or churn, giving you precise insight into future behaviours. Each reader is assigned a unique, customised propensity score, which makes determining the likelihood of a specific action, such as conversion, straightforward.

Propensity modelling may be done in a variety of ways. However, it is mostly through Propensity Score Matching that it can be so reliably predictive. The model compares the actions of former readers and consumers to the actions of new ones. Because groups with comparable tendency ratings copy one another, their future behaviour may be predicted. If new readers have the same behaviour patterns and backgrounds as previous readers, they are likely to behave similarly.


Isn't the problem now solved? Isn't it true that every business should be able to target the appropriate individuals with the right messaging at the right time? That's not entirely true. The reality is that many businesses struggle with propensity modelling, and we've outlined two major reasons why.

The first is that the propensity models that they employ have flaws. For example, the ones you can obtain from your CRM or marketing automation software are scalable, but they're not particularly reliable in terms of the accuracy of the forecasts they can make. In reality, they frequently rely on a restricted set of fundamental characteristics that are generally confined to customer data and campaign-specific transaction history, while ignoring broader transaction history and activity data.

Meanwhile, many internal data science teams generate home-grown variations that aren't always scalable or reliable. The primary flaw in both of these types of propensity models is that they're generally static, which means they don't get more accurate over time when new data is added to them, or adjust to changes in the underlying patterns of that data.

Another issue is that other businesses struggle with implementation. They simply do not adapt their operations to what the models suggest, and as a result, they do not reap the full benefits. Many businesses aren't set up to act on a forecast in real time, provide a dynamic treatment to customers, accurately monitor particular results, or modify predictions and treatments on a continual basis. 

Qualities of Great Propensity Models

A propensity model must be dynamic, productionized, scaleable, and able to demonstrate ROI in order to be genuinely successful. Let's look at each of these characteristics in more detail:

Dynamic- Great models grow over time as new data becomes available, allowing them to become smarter, more accurate, and more in tune with the data's underlying trends. As a result, having a data pipeline and feedback loop that you can utilise to retrain your model on a regular basis is critical.

Productionized- For frequent data intake, retraining and validation, and deployment, dynamic modelling needs a strong data pipeline. Great models will integrate predictions into business processes in a way that makes them intelligible and actionable (often in real time), and they will track and assess model performance over time.

Scalable- Many firms create models for a single campaign and then ditch them. Alternatively, the corporation might devote the necessary resources to developing a new model for each of its campaigns. Neither of these alternatives can be scaled. Effective propensity models must be able to make a high number of predictions and be easily adaptable across comparable contexts in the organisation.

Demonstrate the return on investment- The most effective models go beyond predisposition. Instead, they assist you in determining if the return you'll receive from convincing a prospect to do a desired action is worth the cost. They should assist you in optimising your sales and marketing funnel for maximum efficiency.

Propensity modelling for digital publishers

Digital Publishing | Fastly

Propensity modelling, meanwhile, is a method that may assist digital publishers in converting new customers as well as retaining existing ones.

On the conversion side, propensity modelling analyses the reader's behaviour and feeds them relevant information and offers in order to increase the likelihood of a subscription. For example, if the model determines that a reader is unlikely to convert, they may be given a personalised promotional email with an enticing offer. Customers that are likely to convert and do not require extra persuading, on the other hand, should be presented with an effective paywall right immediately to avoid falling at the final mile. As a result, propensity modelling may be used to determine the most efficient ways to expand your client base, boost revenue, and scale your subscription business without reducing your profit margin through price cuts that aren't essential.

Propensity modelling, on the other hand, looks at how involved people are with your product. Modelling helps your firm to identify which clients are likely to renew their membership and which ones aren't. By concentrating on true churn risks, you can make the most of your time and money. 

Propensity modelling allows you to tailor your strategy to the preferences and needs of each individual reader, rather than using a one-size-fits-all approach. While some readers will subscribe regardless of the number of articles they read or the amount of targeted marketing they receive, others may require distinct personalisation.

Propensity modelling removes the ambiguity from strategy planning by anticipating the activities of both known and unknown readers. With increased trust in client retention and conversion, investments in high-ROI campaigns may be made, resulting in increased income.

Evolok Propensity modelling 

Models & datasets | TensorFlow

We measure every type of user and their level of engagement through machine learning algorithm scores. From their visit history, spend, frequency, devices, favourite section, all of these are used by algorithms to calculate a unique score in real time for every user. These scores indicate a level of engagement of the overall user base and, for each user specifically.

The scores are also captured at different touch points in the users life cycle such as at the point of registration, hitting a number of articles over a certain period and those scoring close to those conversion points, will be challenged or encouraged to convert.

For content owners the benefits are twofold. The paywall becomes automated and dynamic. Therefore removing any manual intervention and allowing a conversion strategy to be truly dynamic. To know more about our service or to book a demo session contact us today. 

Talk to Sales

Sign up for our latest blog updates direct to your inbox