Understanding Customer Churn

One of the biggest challenges of companies across the globe is customer retention. A lot of resources, hard work & planning goes into bringing customers on board & customers’ attrition makes companies do the extra exercise of bringing new customers to fill the void alongside targeted expansion activities. Churn rate analytics is the solution which can prevent this leakage of customer loss.

Consider the situation in the picture. The person is trying to fill the bucket, but there are leakages & hence outflow of water. No matter how much effort is put by the person to fill water either by rapidly filling water in high frequency or in large quantities using larger filling containers, the bucket will not get filled completely. Even if by any means man fails to fill the bucket completely, it will be for seconds since water outflow won't let the filled bucket be sustainable. Now replace man with company, bucket full of water as the target customer base, water inflow be incoming new customers & leakages as consumer attrition, no matter how much effort is put into bringing new customers, customer attrition will never let the company reach its full potential.

What is the churn rate?

The churn rate is defined as the percentage of lost customers or revenue in a certain time period
It is of two types.
Customer Churn Rate=(No. Of customers churned in a period/No. Of customers at the beginning of the period)*100
e.g. - For the telecommunication industry a customer has many options to switch to other alternatives. Suppose a company has seen 1 out of 20 customers canceling the subscription to the prepaid recharge option & switching to other players last year, then the customer churn rate of the company is 5% per annum.

Revenue Churn rate(also known as MRR churn rate/recurring revenue)=(Revenue churned in a period/Revenue at the beginning of period)*100
Revenue churn rate measures the rate at which a company loses revenue due to loss of customers or downgraded subscriptions(i.e subscribing to cheaper services/products offerings by the same company)

Importance of churn rate:

It is useful for a company to compare churn rates with other competitors & see their performance in retaining customers compared to other companies in the same industry. It is important for a company to have a low churn rate because a high churn rate implies high attrition, hence company with a high churn rate needs to spend money to bring new customers on board just to maintain the vacuum created by customers leaving the company. Hence, it is important to understand the churn rate because it impacts the profitability & growth of a company.

Customer Churn Categories:

Contractual churn: Scenario in which customers decide not to carry with their expired contracts with companies, i.e. not to renew services from existing service-providing companies.
e.g. a customer may have subscribed to an OTT platform for 3 months, after 3 months the customer may choose not to renew the subscription for instance due to lack of time to use the OTT platform, hence it comes under the category of contractual churn.

Noncontractual churn: Cases where customers leave potential purchases without completing the transaction. This is highly likely in e-commerce websites or retail stores.
For example, a customer may put the product in a cart & initiate the payment, but at the last moment he might not continue with the payment for reasons like low account balance, skepticism about the product, etc. and this comes under noncontractual churn.

Voluntary churn: This is the scenario where customers cancel their existing contract with the company.
e.g.- A customer who has subscribed for 30 days of internet prepaid services of a company may choose to terminate the contract due to poor network, customer service, or other factors. It is the most harmful form of customer churn for any company.

Involuntary churn: This is the case where a customer willing to continue service offering from the company has to discontinue due to failure of payment for faults on the company’s side like server errors, disabling of payment update system, etc.
e.g-A customer who has defaulted on his credit card payment may not be able to continue his subscription despite his willingness to continue because he can’t afford to pay. It is the least harmful type of churn for any company.

Importance of understanding churn category

Customer churn types may be different but understanding the type helps to know the underlying reason for churn. The cost of acquisition of a new customer is always higher than the retention cost of an old customer hence, it is important for an organization to understand the underlying cause of churn so that it can work on causes that are under its control to rescue customer attrition.

What is Churn Analytics?

It is a technique to measure customer loss rate, understand if a company/product/service is losing customers & if it is losing, reasons for customers quitting. It also helps pinpoint customer segments as well as the timing of their attrition using predictive analysis.

Why Churn Analytics?

The graph above shows the difference in total number of customers of companies with different churn rates over a period of time. As we can see a difference of just 2.5% translates into 4 times more customers for the company having a lower churn rate. This is a snapshot of 80 months of data only, extrapolating this over a longer period will create a gap exponentially between companies having different churn rates. Hence, it is important to minimize churn rate & churn analytics can help understand the underlying causes of churn and will predict in advance the churn of customers helping companies to plan & take action in advance to minimize churn rate.

Benefits of Churn Analytics:

This is the snapshot of one of the dashboards of customer churn. As seen the information provided by it is as follows:

Customer status - in terms of customers who are active, inactive, lost, or new to the company
Churn risk on the basis of income - The risk % of churn of customers on the basis of income can be also easily identified. The company will need to focus only on the specific income groups to prevent churn saving valuable time as well as resources.
Churn risk on the basis of spending as well as the time frame - The bottom left part illustrates the risk profile of customer churn along with the timeline. The segmenting or segregation of customers on the basis of spending groups.
Individual customer churn risk %-As seen in the bottom right part each individual customer churn risk % is mentioned alongside customer details like gender, and spending. It can help the company identify customers at high churn risk % and if necessary can create customized offerings for them.

Let’s look at another dashboard to understand the benefits of churn analytics. Summary of insights a company will be getting from this dashboard created from churn analysis & few of the benefits for the company can be summarised as follows:

The above two illustrations are just the tip of the iceberg. Churn analytics can help companies transform their business into very broad categories by getting deep insights and aligning their strategies to leverage the information obtained from the analysis. In summary, the broad benefits of churn analysis are:
High customer satisfaction- By gaining insights on churn rates across various customer categories, products can be redesigned, pricing can be modified as per customer category & will lead to higher customer satisfaction.

Lower expenditure on marketing- Customer acquisition cost is always higher than retaining old customers. By predicting churn in advance the focus can be shifted to reduce churn and hence extra expenditure on acquiring new customers to fill the loss of lost customers can be reduced.

Low customer acquisition cost- Since analysis will give information about churn of various categories hence, product offerings will be easy to design as per customer segment. It will be easier to attract customers to offerings and extra expenditure on promotional activities will be reduced.

Opportunities for cross-selling & up-selling- Cross-selling is the practice of selling additional products to existing customers. Upselling is the practice of selling high-value products to existing customers. Since, churn analysis gives the churn rate of different categories on the basis of age, income, credit card score, country, etc., this can be leveraged to both cross-sell & sell products that have low churn rates in different customer segments to new segments created from overlapping features of customers(country, age, gender, salary, etc.)

Prevention of loss of market share to competition- Identifying products with high churn rate and mapping with the customer profile where the churn rate is high can be utilized to plan better offerings & hence lost market share to competitors due to high churn can be mitigated.

Industrial applications of churn analysis:

Company: A German bank(name hidden for confidentiality of client)

Background: The financial services sector has increasingly become saturated with a wide array of alternatives available to customers. It has created 2 major challenges
i) Difficulty for banks in differentiating their services from competitors
ii) High churn rate

High churn has resulted in high costs in the form of lost sales

Problem Statement: Due to the sophistication of the datasets, the client, a leading German banking company, found it difficult to determine the reason behind customer churn. The bank was also looking for acceptable models of churn analysis to help them recognize the preferences of various groups of customers as well as identify the associated probability of churn. The client experienced difficulties spanning two main areas, including:
i) Inability to improve the impact of marketing campaigns
ii) Difficulty in analyzing patterns & building effective programs to reduce churn

In order to help them compare scenarios, predict threats, forecast capital, balance risks against projected returns, and work to meet regulatory requirements, the German banking company turned to churn analytics as a tool to solve the issues.
Churn analytics helped banks to:
i) Identify churn reduction opportunities
ii) Align tactical & strategic decision-making for achieving business goals

The final results after analysis & implementation were:
1) Reduction of customer churn from 10% to 3%
2) Improved customer retention rate by 85%
3) Overall annual ROI improved by 70%

When to do churn analysis?

Churn analysis is not a routine job in which analysts or companies do it randomly at any time of the day. Usually, it is done in two scenarios:
1) Red flag- For instance, suppose a company is facing the continuous decline of consumers subscribed to its product/service offering. It will raise a red flag to dig deeper into the reasons underlying this trend.
2) Engagement activity- The reason for doing this analysis may not be bad always. Sometimes companies do it to see the impact of a new initiative on relationships with customers. For example, a company might introduce new engagement activity in the form of thank you emails on birthdays, or post purchase. The company might be interested in knowing its impact on customer attrition & retort to churn analysis and evaluate churn before and after initiation of engagement activity.

Note that little churn is common for any company. Generally acceptable, a churn rate between 5-7% is a healthy churn rate, i.e. not a cause of concern. But the rate may be high & low depending on the industry type Trouble arises when the churn rate is higher than customer acquisition(new customers subscribing). Hence, the following factors indicate the company is having a churn problem & there is a requirement for churn analysis.

Following are the indicators of appropriate time for doing churn analysis:
Churn rate is higher than new customers- This situation leads to a reduction in the customer base and is exacerbated by the absence of upselling(sales technique where an existing customer is sold more expensive products, add, etc to generate more revenue) to existing customers & hence taking a direct hit on revenue of the company.
Shrinking LTV (lifetime value) of customer- The lifetime value of a customer is the overall business worth of a customer to the company over their period of relationship. Constant churning reduces LTV & hence declining LTV requires churn analysis to decipher the reasons.
Churn rate above 10%- As mentioned earlier 5-7% is considered average churn rate, and high or low churn rate value varies across the industry. But the double-digit churn rate is high for any industry and implies something is wrong with either product, sales executive, customer complaint department, or post-purchase service to name a few, and hence makes it difficult for any company to grow in the long run.
More downgrades than upgrades- If the company is offering to add on features and different plans but customers are downgrading i.e. subscribing to cheaper products then it will take a toll on revenue & underlying cause for the downgrading trend has to be identified.

Note: The list above is an exhaustive list of data required but not a complete list. Data requirements will vary depending on the company, industry, macroeconomic factors & discretion of the analyst.

Challenges in churn analytics:

Missing values in the dataset- A data set may have missing values because of the customer not answering every question, the system or staff making errors in collecting data or maybe the variable is not relevant for the customer It has further 2 categories
i) Missing values across a single variable-This is the situation where there are missing values across a data set of only one variable. It can be handled either by coding the missing values with numbers during analysis deleting the missing value data sets or through multiple imputation
ii) Missing values across multiple data sets - The best approach to handle it is to do multiple imputations.

Outliers- Outliers can be identified using descriptive analysis and the reason underlying outliers has to be identified. For example, there might be a sudden spike in the discontinuation of subscriptions with the underlying cause of discount offers from competitors. Identifying the underlying cause will thus help the company focus on its pricing strategy, Outliers can be identified by scatter plots, and line graphs depending on the data type. It can also be identified using statistical methods by standard deviation or Turkey's method(multiple comparison tests to find means that are significantly different from each other).

Method of identification of outliers:

Univariate analysis(analysis of only one variable)
i)Sorting Data:

ii) Heights of 6 people are arranged in descending order & clearly 2 is an outlier Boxplot:

Boxplot shows data distribution by showing quartile ranges. The * mark will help identify the outlier
iii) Bar Graph:

iv) Line Graph:

The outlier would be easily identifiable since it would be far away from the mean value
Bivariate analysis(used to find relationship between 2 variables):
i) Scatter plot analysis: A scatter plot is a type of plot or mathematical diagram to determine the relationship or patterns between two variables by plotting individual pieces of data in the form of data for two variables

Here the relationship between output & input is being analyzed. For the point marked under the circle, the output is high & simultaneously input is low compared to other data points hence it is an outlier.
Multivariate Analysis(It is analysis of 3 or more variables):
i) Tukey's test: This test compares the differences between means of values rather than comparing pairwise values(done previously with casewise significance chosen)

-Outliers due to data entry errors can be identified by cross-verifying the outlier's original value by either interviewing the customer or checking the original data sheet & it should be deleted or corrected by going to original source of data to get the correct value
-Sampling error in the form of selecting not a sample representative of the population can also cause outliers & hence these outliers should be deleted from the data to be analyzed. e.g -Selecting people of age groups between 20-30 for study from a population where age varies from 0-80 years would lead to sampling error
-Outliers due to natural variation are not unusual & hence they need not be modified or deleted from the dataset. e.g.- Consumer demand may become 10 times during a particular time period due to seasonality, and festivals. So, these outliers don’t need to be removed because they are naturally occurring

Churn Analytics Methodology:

There are multiple techniques to do churn analysis among which popular ones are:
1) Logistic regression-It is the most popular technique, due to its reliability and quick outcome. It gives binary dependent variables as output & independent variables can be multiple.e.g-Logistic regression to check whether a customer will inscribe or not in the next 3 months with X1, X2, and X3 as independent variables which will include customer demographics data, transaction data, complaint issues,
INPUT-OUTPUT
X1
X2 --> Logistic regression--> 1 or 0
where the 1-customer will unsubscribe
0-customer won’t unsubscribe
Note - For output variables that are not binary, they can be transformed into categorical or ordinal categories

Categorical variables- These are variables that have categories among themselves but no inherent ordering between them. For example, the population can be categorized into various age groups say 0-20 years,21-40 years, 41-60 years, and 61-80 years for analysis of consumption of internet data per day. As we can see, the categorical variable of age group is created by segregating the population into age groups & there is no inherent ranking in the age group.

Ordinal Variables- Similar to a categorical variable is an ordinal variable. The distinction between the two is that the categories have a simple ordering. For example, suppose you have an economic status variable with three categories. Besides being able to classify individuals into these three categories, the categories may be ordered as low, medium, and strong.

2) Cox regression- Cox regression (or proportional hazards regression) is a way of investigating the effects of various variables when a single occurrence takes place. The aim of the model is to assess the influence of several factors on survival at the same time. In other words, it enables us to analyze how defined variables affect the rate of occurrence of a specific event (e.g., illness, death) at a specific point in time. This amount is usually referred to as the danger intensity. The aim of the model is to assess the influence of several factors on survival at the same time. In other words, it enables us to analyze how defined variables affect the rate of occurrence of a specific event (e.g., illness, death) at a specific point in time. This amount is usually referred to as the danger intensity. In the literature on survival analysis, predictor variables (or factors) are typically called covariates. In short, it is possible to view the hazard feature as the chance of dying at t-time.
For churn analysis, the churn will be a hazard feature & the factors included in the hypothesis will be covariates.
Both methods are extensively used & very reliable. Selection of method depends upon data type & output requirement.

Conclusion

Churn Analytics is the solution for identifying reasons for customer attrition/churn. It can pinpoint the underlying factors behind the churn & hence will provide the decision makers a direction to devise the strategy for reducing churn Churn is one of the major reasons behind lowered profit henceforth low ROI of the company. The analysis will hence improve profitability alongside customer satisfaction and lower customer acquisition costs. This can be applied across a broad range of industries in an environment where increased competition has reduced the perceived differences in product offerings.