It’s pretty simple: churn happens when your customers are customers no longer. For any business (even those gaining customers quickly), this can be a devastating problem; but fortunately predictive analytics can help anticipate and mitigate this loss.
Predictive analytics might sound intimidating, but the reality is that churn prediction is relatively basic and actually very accessible to even small or less experienced teams, so the hardest part is just getting started.
The basic requirements and principles mirror the 7 Fundamental Steps to Complete a Data Project.
This blogpost will skim the surface on how to begin a churn prediction project, but if you want an in-depth walkthrough of the entire process beyond just getting started with additional detail for both beginners and more advanced readers, check out the full guidebook.
Define Churn
The first, and perhaps most important, aspect of predicting churn is defining it. This can be more difficult than it seems if your business is not subscription-based. Customers can go for a long time without making a purchase and suddenly buy again, or they can taper off slowly over time. Ask yourself:
- Who exactly should be considered in the definition of customer when considering churn? Someone who purchased only one time? Twice? Over what time period?
- At what point is a customer considered no longer a customer?
- When is (s)he considered at risk of churning?
- When is the optimal time in customers’ life cycles to try to re-engage them?
Of course, this definition might need to be adjusted or tweaked during the course of the project depending on your findings. You may start out with one definition but find further along in the project that it’s too broad or too narrow. So it’s worth spending time and resources doing initial analysis on these questions to come up with a strong definition from the start to avoid having to rework the definition later.
Choose Clear Project End Goals
Starting a churn prediction project without clear goals about how those predictions will be used can ultimately prove to be a waste of time for both data teams and marketing or business teams.
At this stage before even starting to work with data or develop models, it’s critical to ask the question: what will churn predictions be used for? This question is not so obvious, as churn prediction can have many ultimate goals and depends highly on the business doing the prediction. For example, predictions can be used for:
- Short- term solutions like marketing campaigns to re-engage likely churners.
- For example, email campaigns or special offers sent at a time when a likely churner is considered at risk.
- Uncovering potential deeper drivers of churn that can be addressed long term.
- For example, maybe there is an issue with the product that is blocking customers’ ability to come back easily or there are in-product improvements (or potential new features) to be made to prevent attrition.
Whatever the goal, the marketing, business, or product teams must be clear from the start about expectations for the project, what they will do with the results, and any specifications for how the predictions should be delivered so that the ultimate project goal can be realized without additional roadblocks.
Start Working with Data
Only once churn itself as well as the project’s goals have been properly defined is it time to actually move on to the meat of the churn prediction project: working with data to identify customers likely to leave. The next steps are:
- Choosing the data to work with
- Exploring, preparing, and enriching that data
- Applying predictive models
- Visualizing the data
- Iterating on the results and deploying them to production