Case Study - Predicting Credit Card Attrition

Situation

The Credit Card area of a large retail bank wanted to maintain - and increase - the profitability of their credit card portfolio and ensure they were not losing their good customers. They recognised it was more cost effective to spend marketing resources retaining existing customers than to spend large amounts marketing for new customers.

Vision

One of the vital elements in a relationship management toolkit is a measure of how likely a client is to close their credit card account. The vision was to combine this measure with metrics of current and future profitability of the customer to the bank.

Then when a customer was at risk of leaving the bank, a decision could be made as to the optimum marketing spend to be applied to retain this customer.

Specific Requirement

The bank required an alarm to be generated when a customer appeared to be thinking about closing their credit card account. This alarm was required with adequate time before the account was actually closed so that action could be taken to intervene and encourage the customer to remain with the bank. This required a rigorously defined, actionable and timely metric of likelihood of attrition to be generated for all credit card customers to be used as the basis for retention strategies.

We Provided

Pre-analysis that revealed some previously unknown and sometimes counterintuitive insights into the portfolio as a whole as well as between brands, vintages and card types.
A summary of the factors which were most influential in increasing or reducing the chance of attrition.
A propensity model predicting the likelihood of attrition in the next three months.
An attrition score from 1 to 1000 used to rank customers from lowest to highest risk of closing their account.
A simple tool for estimating dollar savings or increased response rates from targeted retention campaigns using the attrition score.
Full documentation of the whole project including modelling methodology, diagnostics and results.
End-to-end SQL scripts for monthly runs of the scoring process, including comments and built in quality checks.

Result

The propensity model had a 60% success rate in the top 1% ranked as about to close their account. The bank now has a process to identify customers who are about to close their account, weigh up the amount of marketing spend that should be applied to retain the customer – and all with enough time to intervene to convince the customer not to close the account.

Knowledge was gained by the bank regarding the factors that influence a customer to leave – knowledge that can be used when planning future campaigns. All this results in reduced marketing spend in retaining the best credit card customers and increased profits.