How Does Machine Learning Support Subscription Ecommerce?
More and more native ecommerce companies are moving towards adopting the subscription model. It’s important to understand how machine learning and predictive insights can help marketers of subscription ecommerce businesses.
Machine Learning and Subscription Cortex
Our lifecycle marketing product Subscription Cortex harnesses the power of AI for ecommerce marketers, and this article covers various machine learning approaches and answers important questions for marketers.
Subscription Lead Scoring: Which acquisitions/leads are good, and which are bad?
Subscription businesses typically spend a considerable amount of money acquiring users from social media, referral programs, freemium models, etc. If the users don’t maintain their subscription for at least a few cycles, it can be highly detrimental to the sustainable growth of the company.
From a machine learning standpoint, identifying good vs. bad users is a good fit for binary classification algorithms. Given all the information available about our users at signup, a classifier model can learn which features are associated with long-term customers and which are associated with churners (for a more detailed description of our analog for non-subscription businesses, see our posts about WPP). For example, let’s suppose we have two users: Jennifer, a 31-year old who resides in LA, was acquired through a Google ad via an iPhone. Jennifer could be a high intent subscriber. On the other hand, Derek, who is of unknown age and resides in New York, was acquired through Facebook via a PC, could be a low intent subscriber (see Fig 1.1).
This segmentation on acquisition can be a powerful way to deploy or test various strategies (see Fig 1.2). Understanding which locations, registration sources, and personas are your cash cows is one of the foundations of building a strong business model.
Subscription Cancellation Prediction: Which subscribers are about to cancel?
Most marketers agree how hard it is to win back canceled subscribers and how high the associated costs are. Having a forewarning on users about to cancel their subscription is a vital advantage in a modern marketer’s toolkit. Marketing can be proactive instead of reactive. Subscription pauses or cancellations reduce the lifetime value of users and bring a sudden loss in revenue.
How do you have a forewarning on “At Risk” users? User online behavior, subscription parameters, and product responses from users can be used to determine how well they are doing in the subscription cycle. At ReSci, we currently track about 80 user features for this purpose. These features are passed into a custom clustering model to identify an “At Risk” cohort which can be used for targeting marketing campaigns. In addition, they’re also updated on a daily basis and react to every action made by the user.
Subscription CLV: What’s the lifetime value of various personas?
Our earlier blog post talks in detail about how machine learning can be used to predict lifetime spend or future spend for a given user.
Once a predictive system is put in place, it’s important to analyze the acquisition sources, locations, and demographics with high or low CLVs. ReSci’s Cortex provides a complete 360° view (Fig 1.4) of your high- and low-CLV segments and updates it every day.
A newfound application of the CLV prediction is to leverage the High CLV segment to create “lookalike” Facebook audiences. This can lead to better customer acquisitions and reduced costs to achieve greater lifetime value for users. The outcome is better acquisitions results in higher retention rates and longer subscription cycles.
Engagement Scoring: Which current subscribers are Engaged vs. Passive?
Subscribers are typically passive or engaged. From the above example, you can get an idea of some characteristics of these subscribers.
Machine learning can be used to segment users into the above mentioned categories. Features shown below can be used in a classifier to obtain the 2 segments.
- email activity
- page visits
- session duration
- browsing behavior
Experiments concluded that engaged users are 3X more responsive as compared to passive users.
Conclusion
About The Author
Vedant Dhandhania is a Machine Learning Engineer at Retention Science. He helps predict customer behavior using advanced machine learning algorithms. His passion lies in the intersection of Signal Processing and Deep Learning.
Take a peek at some of our resources below to drive you towards success on your marketing journey!
