Because many researches proved that it’s cheaper to retain your customers than acquire new ones, customer retention became one of the most critical goals for businesses. Churn itself is a term that describes a phenomenon of customer retention. With the churn analytics you are able to predict who and when become less and less engaged and they are able to be lost. This feature lets you find answers on the following questions:
- Why do customers leave?
- Who is likely to leave?
- When are clients likely to leave?
- Users of what kind of products are prone to churn?
- What kind of actions need to be taken to prevent churn?
In the Synerise platform users can find churn analysis in two places:
- On the customer’s profile in CRM to get information about the probability of a churn for an individual customer.
- In the dedicated Churn anlysis section in the Analytics module to get information about the retention statistics in general.
- Historical transaction data (minimum 3 month history).
- History of page views (optional)
- Churn analytics needs to be switched on by Synerise as well
- Contact the Synerise support and request switching on the Churn analytics feature
Churn analysis in CRM
The Synerise users have a possibility to check the probability of churn for their every customer. The Churn section is available on the customer’s profile in CRM.
Churn is calculated on the basis of top indicators, which are determined individually with the assistance of Synerise support. Users must also consult the preferred time intervals in which the top indicators are calculated.
In the agreed time intervals, a
churn.scoreevent is generated and it contains details about the top indicator results. The results for top indicators are given in percentage points and they can take both negative and positive values.
The score below Churn presents the increase (red value) or decrease (green value) in the churn risk since the last calculation (in percentage points as well).
Churn analysis in Analytics
Go to Analytics > Churn Analytics.
SHAP values provide explanations of the output of machine learning models by breaking down a prediction in order to show an impact of each feature.
- This is the list of features. Such phenomenon as churn is shaped by many features which usually are described as single measurable properties of phenomena such as churn.
- Importance - A value that presents a relative change of the churn score if the feature value changes. The higher value of importance means that if a value of feature changes, the churn score will be bigger on average.
- Impact on model output - The plot shows SHAP values of every feature for every sample. Features are presented with regards to their magnitude on churn prediction. High values of certain feature are marked as red-shaded dots, whereas these with low values as blue-shaded.
- Average impact on model output magnitude - The mean absolute value of the SHAP values for each feature that has impact on churn prediction.
- Importance - The importance of chosen feature
- Feature dependency - The scatter plot shows dependencies between SHAP values of certain feature and values for each observation for the same feature. Dots are colored with regards to the value of the another feature chosen by user from dropdown menu at the left-top of the frame. It is particularly useful in highlighting interactions between two features. The strength of these interactions is explained by metric
Feature interaction importance, in which the higher number means the greater impact on the churn prediction.’
- Statistics - The box-plot depicts the distribution of a certain feature in a standardized way, where bottom edge of the rectangle is equal to 25th percentile (Q1) of the feature values, line in the middle is median of feature values, the top (Q3) is nothing else as 75th percentile. Minimum and maximum values are presented as whiskers added to the rectangle. Together with box-plot we present also the basic statistics such as mean feature value and its standard deviation.
- Histogram - The chart presents the distribution of the specific feature, together with its cumulative distribution and quartiles Q1, Q2, Q3.
A cohort analysis divides customers who subscribed in a particular period into groups (cohorts). Due to this division you can analyze their engagement and how it has changed over at the same time being unaffected by the individuals in other groups.
- Initial step date - This is the starting date of the analysis.
- Customers - In the example presented in the illustration, each month starts with various number of customers.
- % of returned customers relative to the first month - This column presents the number of people who returned in subsequent months compared to the number in initial month (1). The retention of customers who joined in a particular month and the behavior of this group (whether customers stayed or left) in every month is presented in verses. For example, in February 2018 the initial number of customers amounts to 33 129. However, in the next month 99,66% of the original joiners returned, two months later this number diminished as 87,50% of the original group returned, etc.