In our last post, we discussed the data identification and research phase of the churn model development. We started with collecting the right data (part 1), and then moved onto asking the right questions (part 2) to find conclusions about our churn triggers. Now, we reach the next step which is creating the model and putting it to work.
The critical requirement for this step is the understanding you developed in your research in part 2 of this post series. The last step of making the model requires your data set and understanding to be grounded in some conclusive concepts to give you a starting point. Once you have that, you can use the model for retention tracking and optimization campaigns.
Calculating Lift and Drop
The outcome of a good model is to find triggers and behaviors that equate to lift, or increased retention, and drop, an increase in churn. Each of your attributes can be mapped to a lift or drop quotient using an equation. We’ll use the table below to better understand how to create a lift or drop quotient for each trigger.
The hardest part about calculating lift is the definition of the control group. There are many ways you can do this, but for our example we are going to use the number of users with the exact same attribute sets, or triggers, except the one trigger we are trying to isolate for- our calculation of lift or drop. In our example below, we have a control group and a test group.
|Group||Total Users||Churn Users||Churn Rate|
|Trigger Test Group||2000||250||12.5%|
From this control group and trigger group, we can see there is a potential for an 11.2% lift in customer retention for that particular trigger. Of course the only way to validate that is to make sure there are no other triggers that separate those two groups. If there are, you are not doing analysis on a particular trigger, but the accumulation of a set of triggers. There is a big difference.
You can take lift and drop to the next level by comparing it to your LTV, or lifetime value for your customers to get profitable measure to your efforts. For example, if each customer had a $100 value, adding 11.2% of 6M users would be $67M in additional revenue.
You can take the same steps to find items that contribute to drop. In these cases, you might not have accumulated data such as page views or time on site. You might have to use computed data, such as time between visits or failed funnel activities to garner real insights into the negative behaviors.
Putting it to Work
Your model will have to be used and validated on a consistent basis. In whatever interval you calculate your churn, daily, weekly or monthly, you’ll need to setup a churn analysis schedule to do the same. We recommend taking a longer view on your data, since many factors can change based on volume of users. The next step once the model is in place is to start with your easiest “wins” for behaviors modification and retention campaigns. There are some things that you can not teach or train, some you can. Either way, having a churn model will help you plan your customer retention efforts and give you a concrete measurement of your success. Good luck!