To compete and stay relevant in a rapidly changing world, businesses need to embrace the idea of using data and AI for their marketing activities. This also means a changing dynamic in the activities and skills that marketers will need to have and to execute in order to implement marketing AI solutions. But it doesn't mean all marketers need to become data scientists overnight or vice versa - instead, the focus should be on collaboration.
To realize this goal, we suggest setting up a use case that promotes cooperation between colleagues with different skill sets. One of our favorite cross-team approaches between marketing and data science is to practice a use case involving churn analytics.
Churn (or attrition), put in the simplest terms, is when customers leave and stop buying your product or using your service during a defined time frame. In order to do keep churn rates as close to zero as possible, companies in nearly every industry need to treat churn as a top priority. If ignored, churn can plateau the growth of any business, even the ones that can gain customers quickly.
Churn Detection Depends on the Subscription Model
The first step in a churn-based data science project is to define the model used by an organization. Typically, there are two varieties: subscription and non-subscription churn.
Subscription churn happens in businesses where users or customers are on contract for a set period of time, and customers cancel or choose not to renew their subscription after that contract is up.
Non-subscription churn happens when users or customers can end their relationship with your business at any time - they come and go at will. A customer may gradually over time reduce their purchase frequency, or they may all of a sudden never buy again.
Determining whether or a not a customer will become a churner (i.e., no longer remain a customer) is fairly straightforward in subscription models, but a bit more challenging in non-subscription models. In subscription models, a customer churns when they request cancellation of their subscription. In non-subscription models, however, you need to analyze your customer’s behavioral tendencies in order to identify potential churn (e.g., the amount of time since he last used the company’s services/products).
Non-subscription churn requires collaboration among several teams to predict accurately - the marketing side generally defines what constitutes churn for their business (lack of action after weeks, months, or years), and then it’s a back-and-forth, iterative process with data teams to arrive at the right model.
Using Machine Learning to Model Churn
Churn prediction projects are typically launched when the customer acquisition rate diminishes. For most companies, the customer acquisition cost (cost of acquiring a new customer) is higher than the cost of retaining an existing customer, sometimes by as much as 15 times more expensive. Therefore, the challenge of implementing a successful churn project is to increase customer loyalty and, consequently, increase company revenue.
There are two complementary modeling approaches used to predict churn:
- Machine learning model (short term action). Develop a machine learning model to analyze performances that will enable short term actions. Based on the outcome of this analysis, our clients are able to undertake one-shot actions to reduce churn;
- Analytical model (long term study). Develop a model to understand the reasons causing the churn. This deeper knowledge enables our clients to attack the root of the problem and to understand how to reduce churn.
In both cases, it is crucial to connect your models to relevant short or long term marketing-driven actions in order to attain churn reduction. In both modelling and marketing strategies, only a combined approach of mixing short term with longer term approaches will have an effective and sustainable impact on churn reduction.
Creating a Churner Profile and Identifying Churn Behavior
Churn analytics projects can be addressed by data science and marketing teams thanks to machine learning modeling (classification) with a defined target. The target is known in subscription business models while it needs to be defined in non-subscription scenarios. There are several considerations to take into account when identifying churn behavior:
Segment your customers based on their behavior and address the question, “Which customers do we care about?” Only the best? The most valuable? Regardless of the answer, a churn reduction campaign should be targeted toward a well defined customer segment;
Compare to Control Population
By understanding the variations in churn behavior, new customer classes can be created and refined. On one end of the spectrum, there are customers who interacted with the product at least once, but no longer visited afterwards. The other end of the spectrum includes customers who make frequent uses or purchases and are heavily engaged with the product. In this context, the definition of a “new customer” can be formulated along with an understanding of customer groupings;
What Makes your Churner Different?
Data collected from the above analysis, when subjected to machine learning modelling, enables your company to discover differential patterns among churners and identify what makes your churners different from others.
Implementing the Churn Scoring Mechanism
Churn scoring is used to assign a score to customers that conveys the potential loyalty of the customer. Implementing a churn scoring mechanism relies on a pair of processes:
- Find relevant features: Customer features, such as social information and behavior-based actions, are used to paint a picture of who your customers are. Start the churn scoring process by finding the customer features that are the most relevant to your churn calculations;
- Compute a churn score: Churn score computation combines all relevant customer features to determine exactly how likely specific customers are to abandon your product/service. At this point, machine learning technology takes over — predictive algorithms are fed into Dataiku and the best one is selected. It is then deployed to calculate a churn score.
The process of churner identification and behavioral analysis involves expertise from both data scientists and marketing specialists: one party understands the customers whilst the other can measure & analyze behavior. At the end of this process, it is time to apply the information learned to the company’s loyalty program. The output of a churn project is a dataset that contains the customer ID and an associated churn score. This churn score indicates the probability of the customer abandoning your product or service.
Churn scores enable data science and marketing to build business rules together in order to define customer segments. In order to achieve optimal results, the actions need to be customized based on your business requirements and your knowledge of customer behavior and expectations.
Finally, in order for your churn prediction project to be successful, you want to avoid one common pitfall: perceiving the model as a one-shot study. You won’t do much with a single iteration of a churn model and you risk overlooking critical aspects in the project, such as scalability (applying it to grow larger customer bases) and reproducibility (replicating the same approach in future). Multiple iterations help you create a successful scalable and long-term strategy to retain and increase customer loyalty.