Pre-delinquency modeling to reduce bad debts

Client:

Large International bank with more than 800,000 active customers

Business Problem:

Reducing customers with delinquent loans with marketing analytics

Solution delivered:

Data-driven campaign flow design for different customer segments based on the predicted propensity of late payments resulting in 2% reduction in delinquent customers

Tools used:

1. R

2. MySQL

Non-Performing Loans (NPL) can lead to disastrous losses for banks. Clearly, the profits on good loans do not balance out the the losses on bad loans and so the NPL should be ideally 0 (realistically, as low as possible).

My client, a large international bank, faced the pervasive problem of delinquent loans. A traditional collections service would deeply cut the bank's profit margins and so I proposed to go down the path of marketing analytics.

Most people are generally responsible for their monthly payments and while the goal is to reduce NPL, it is important to note what the solution is NOT:

1. to use the FICO score card (pay on time or your credit score will drop so low that you will never ever get a loan),

2. to scare customers with a harsh message (give me your money or I will kill you and your family, well maybe, not that harsh!),

3. send repetitive reminders, and

4. use any language that could potentially affect their relationship with their customers

The problem was not solvable with just propensity modeling where a probability score card could help target high-risk customers. The solution required a more holistic approach; a careful design of customer touch points, appropriate content and timing of the communication, and the response of customers to foster a conversation than mere reminders.

It was one of the most interesting data-driven experimentation that I have worked on in my career where I proposed the solution to try out different campaign automation flows to compare their effectiveness and assess what works best for the bank.

A random stratified sample that includes customer demographics and loan details was used to test different campaign flows based on the propensity of customers.

The final automation flow resulted in a 2% decrease in new delinquent accounts baselined with the traditional method of execution.