This is the second article of the series on Predicting Customer Churn using Machine Learning and AI. In case, you missed the first here is the link.

In the earlier post, we discussed our proposed approach which consists of six steps as given below:

  1. Business Understanding
  2. Data Understanding
  3. Data Preparation/pre-processing
  4. Modeling
  5. Evaluation
  6. Real-world Validation and Deployment

1- Business Understanding

For the sake of case study, we are assuming a dummy Fintech company who is providing digital wallets to it's customers. Let's try to understand the business. Assume that company has a reasonable customer base but also looking to increase the revenue.

Acquiring New Customer vs. Retaining Existing

If you want to increase profit, one way is to increase customers. Find new customers who don't have your product and try to on-board them. Now-a-days, markets are saturated and finding a brand new customer is nearly impossible. Also, taking them away from their current provide is also very difficult. Retaining a customer through an offer is often cheaper if we already have a sense that he is going to leave us.

Benefits of retaining an existing customer

Generally, cost of acquiring a new customer is 5-6 times more than retaining a existing customer. Cost Benefits are mainly because of low maintenance cost. Associations with activation, pin setups, automatic payment settlements are very less. Other benefits includes Loyalty benefits, e.g. Longer you have customer, less likely they will leave the company. Also, satisfied customers are likely to bring friend and family. Free advertisement!

Understanding Type of Attrition

Voluntary - the Customer decides to quit and switch to another service provider.

  • Type-1: Dissatisfaction with the quality of service. May be due to Not fulfilling service level agreements, too high costs, not competitive price plans, no rewards for customer loyalty, no understanding of the service scheme, bad support, no information about reasons and predicted resolution time for service problems, no continuity or fault resolution, privacy concerns, etc.
  • Type-2: Quits without the aim of switching to a competitor. Reasons could be the changes in the circumstances that prevent the customer from further requiring the service, e.g. financial problems, leading to impossibility of payment; or change of the geographical location of the customer to a place where the company is not present or the service is unavailable.

Involuntary - the Company discontinues the contract itself. Main reason for such attrition is the inability of customer to pay which leads to delinquency and ultimately the cancellation of card.

Never Used Never Paid - these are customers that have acquired the services but have never utilized. Hence never made payment as well.

Inactive Customer - there is no activity on Customer account for six months. Customer has not made a transaction (spend or payment) in last six months. Such customers are also sometimes considered as attrited customers.

So, where should be the focus?

In conclusion, Voluntary attrition is quite hard to predict. Out of two types of Voluntary attrition, Type-2 attrition takes a very small fraction of overall voluntary attrition. Therefore, the focus should be more on Type-1 attrition. Business are really interested to react positively by taking appropriate action to prevent Type-1 attrition. So, to prevent Voluntary attrition, we must know

  • who are the possible attrition accounts with a low probability of error in the prediction.
  • why this specific customer has decided to leave the company for the benefit of a competitor.

To prevent Voluntary Attrition what we should do?

Can we use Machine Learning to intelligently predict the unknown future to us? Yes!!! ... but wait how does machine learning work? Collect a ton of data and analyze it! Easy right? Not so fast. Some says, there are tools in market now!

Yes, there are advance tools in the market, But don’t think that you just have to provide data and they will do magic for you. There is a lot of science and statistics behind it that you must know before you could use any of the algorithms provided by such advanced tools.

In machine learning and statistics, classification is a supervised learning approach in which the computer program learns from the data input given to it and then uses this learning to classify new observation. This data set may simply be bi-class (like identifying whether the person is male or female or that the mail is spam or non-spam) or it may be multi-class too. Some examples of classification problems are: speech recognition, handwriting recognition, bio metric identification, document classification etc.

Customer Attrition can be treated as a Classification problem in Machine Learning.

In the next post of this series, we will look at the second phase i.e. Data Collection

until then...

Cheers :)